Perform additional reliability testing for endorsement review.
Percentage of surgical inpatients who experienced a complication and then died within 30-days from the date of their first “operating room” procedure. Failure-to-rescue is defined as the probability of death given a postoperative complication.
Measure Specs
- General Information(active tab)
- Numerator
- Denominator
- Exclusions
- Measure Calculation
- Measure Record
- Point of Contact
General Information
N/A as this is not a paired measure.
Website URL not available; Final measure specifications for implementation will be made publicly available on CMS’ appropriate quality website, once finalized through the CBE endorsement and CMS rulemaking processes.
Medicare inpatient claims data, including Medicare Inpatient Encounter (shadow billing) data for Medicare Advantage enrollees, in combination with validated death data from the Medicare Beneficiary Summary File or equivalent resources. CMS receives death information from a number of sources. The main sources CMS uses to develop its death information are Medicare claims data from the Medicare Common Working File (CWF), online date of death edits submitted by family members, and benefit information used to administer the Medicare program collected from the Railroad Retirement Board (RRB) and the Social Security Administration (SSA). Overall, over 99.9% of death days have been validated. As for other CMS 30-day mortality measures, the "Valid Date of Death Switch" is used to confirm that the exact day of death has been validated.
Numerator
Patients who died within 30 days from the date of their first “operating room” procedure, regardless of site of death.
Number of verified deaths (STUS_CD=20) within 30 days from the date of the first eligible operating room procedure (Table 1), regardless of site of death, among discharges meeting the inclusion and exclusion rules for the denominator.
This measure uses submitted claims data and vital status data from the Medicare Beneficiary Summary File (or equivalent resources, such as a Vital Status File) to calculate the measure score. All data elements necessary to calculate this numerator are defined with the attached technical specifications.
Denominator
Patients aged 18 years and older admitted for certain procedures in the General Surgery, Orthopedic, or Cardiovascular Medicare Severity Diagnosis Related Groups (MS-DRGs) who were enrolled in the Medicare program and had a documented complication that was not present on admission.
Documented complications include: cardiac events, congestive heart failure, hypotension or shock or hypovolemia, pulmonary embolus or deep vein thrombosis or phlebitis, cerebrovascular accident (CVA) or transient ischemic attack (TIA), coma, seizure, psychosis, nervous system complications, pneumonia or pneumonitis, pneumothorax/effusion, respiratory compromise or bronchospasm, internal organ damage or perforation, peritonitis, gastrointestinal bleed and blood loss, sepsis, deep wound infection or wound complication, renal dysfunction, gangrene/amputation, intestinal obstruction or ischemia, retained foreign body, pressure injury, orthopedic complication, hepatitis or jaundice, pancreatitis, necrosis of bone (thermal or aseptic), osteomyelitis, disseminated intravascular coagulation (DIC), pyelonephritis, or other postsurgical complication.
DENOMINATOR OVERALL
Discharges for patients ages 18 through 89 years with any listed ICD-10-PCS procedure code for an operating room procedure (Table 1) and all of the following:
-Enrolled in the Medicare program
-Any admission type in which the earliest ICD-10-PCS code for an operating room procedure (Table 1) occurs within the qualifying period, starting three days prior to the date of admission and ending at the date of discharge
-Meet the inclusion and exclusion criteria for one of the denominator complication categories (Tables 3-5)
And meeting one of the following criteria:
-Eligible discharges assigned to the General Surgery, Orthopedic, or Cardiovascular Medicare Severity Diagnosis Related Groups (MS-DRGs: Table 2)
OR
-Eligible discharges assigned to the ECMO or Tracheostomy Medicare Severity Diagnosis Related Groups (Table 2; MS-DRGs 003 or 004), and
- with an MDC for diseases and disorders of the circulatory system; digestive system; hepatobiliary system and pancreas; musculoskeletal system and connective tissue; skin, subcutaneous tissue and breast; or endocrine, nutritional and metabolic diseases (Table 2; MDCs 05, 06, 07, 08, 09, 10), and
-with any listed ICD-10-PCS code for a procedure assignable to MS-DRG 003 or 004 (Table 1; FTRPXCHGTOMSDRG003004P), that, in the absence of a code for ECMO (Table 5) or tracheostomy (Table 6), would assign the discharge to a denominator eligible MS-DRG (Table 2), and
-without any listed ICD-10-PCS procedure code for ECMO (Table 5), and
-without any listed ICD-10-PCS procedure code for tracheostomy (Table 6) occurring before or on the same day as the first non-tracheostomy operating room procedure
Denominator Category 1_Cardiac Event
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for cardiac event not present on admission (Table 3; FTR1CARDEVENTD) or any listed ICD-10-PCS procedure code for cardiac event (Table 4; FTR1CARDEVENTP) at least one day after the first qualifying operating room procedure (Table 1)
Denominator Category 2_Congestive Heart Failure
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for congestive heart failure not present on admission (Table 3; FTR2CHFD)
Denominator Category 3_Hypotension/Shock/Hypovolemia
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for hypotension, shock, or hypovolemia not present on admission (Table 3; FTR3SHOCKD)
Denominator Category 4_Pulmonary Embolus/Deep Vein Thrombosis/Phlebitis
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for pulmonary embolus, deep vein thrombosis or phlebitis not present on admission (Table 3; FTR4PEDVTPHD) or any listed ICD-10-PCS procedure code for pulmonary embolus, deep vein thrombosis or phlebitis (Table 4; FTR4PEDVTPHP) at least one day after the first qualifying operating room procedure (Table 1)
Denominator Category 5_Cerebrovascular Accident (CVA)/TIA
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for cerebrovascular accident or transient ischemic attack not present on admission (Table 3; FTR5CVAD)
Denominator Category 6_Coma
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for coma not present on admission (Table 3; FTR6COMAD)
Denominator Category 7_Seizure
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for seizure not present on admission (Table 3; FTR7SEIZD)
Denominator Category 8_Delirium/Psychosis
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for psychosis not present on admission (Table 3; FTR8PSYCHD)
Denominator Category 9_Nervous System Complications
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for nervous system complications not present on admission (Table 3; FTR9NERVSYSD)
Denominator Category 10_Pneumonia/Pneumonitis
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for pneumonia or pneumonitis not present on admission (Table 3; FTR10PNEUMOD)
Denominator Category 11_Pneumothorax
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for pneumothorax not present on admission (Table 3; FTR11PTXD) or any listed ICD-10-PCS procedure code for pneumothorax (Table 4; FTR11PTXP) at least one day after the first qualifying operating room procedure (Table 1)
Denominator Category 12_Respiratory Compromise/Bronchospasm
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for respiratory compromise or bronchospasm not present on admission (Table 3; FTR12RESPCOMPD) or any listed ICD-10-PCS procedure code for respiratory compromise/bronchospasm (Table 4; FTR12RESPCOMPP) at least one day after the first qualifying operating room procedure (Table 1)
Denominator Category 13_Internal Organ Damage/Perforation
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for internal organ damage or perforation not present on admission (Table 3; FTR13ORGDAMD) or any listed ICD-10-PCS procedure code for internal organ damage or perforation (Table 4; FTR13ORGDAMP) at least one day after the first qualifying operating room procedure (Table 1)
Denominator Category 14_Peritonitis
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for peritonitis not present on admission (Table 3; FTR14PERITD)
Denominator Category 15_GI Bleed and Blood Loss
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for gastrointestinal bleeding or blood loss not present on admission (Table 3; FTR15GIBLEEDD)
Denominator Category 16_Sepsis
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for sepsis not present on admission (Table 3; FTR16SEPSISD)
Denominator Category 17_Deep Wound Infection/Wound Complication
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for deep wound infection or wound complication not present on admission (Table 3; FTR17WOUNDD) or any listed ICD-10-PCS procedure code for deep wound infection or wound complication (Table 4; FTR17WOUNDP) at least one day after the first qualifying operating room procedure (Table 1)
Denominator Category 18_Renal Dysfunction
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for renal dysfunction not present on admission (Table 3; FTR18RENALD)
Denominator Category 19_Gangrene/Amputation
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for gangrene or amputation not present on admission (Table 3; FTR19GANGAMPD)
Denominator Category 20_Intestinal Obstruction/Ischemia
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for intestinal obstruction or ischemia not present on admission (Table 3; FTR20INTOBSTISCHD)
Denominator Category 21_Foreign Body
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for foreign body not present on admission (Table 3; FTR21FORBODYD)
Denominator Category 22_Pressure Injury
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for pressure injury not present on admission (Table 3; FTR22PID)
Denominator Category 23_Orthopedic Complication
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for orthopedic complication not present on admission (Table 3; FTR23ORTHOCOMPD) or any listed ICD-10-PCS procedure code for orthopedic complication (Table 4; FTR23ORTHOCOMPP) at least one day after the first qualifying operating room procedure (Table 1)
Denominator Category 24_Hepatitis/Jaundice
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for hepatitis or jaundice not present on admission (Table 3; FTR24HEPATD)
Denominator Category 25_Pancreatitis
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for pancreatitis not present on admission (Table 3; FTR25PANCD)
Denominator Category 26_Necrosis of Bone (Thermal or Aseptic)
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for necrosis of bone (thermal or aseptic) not present on admission (Table 3; FTR26NECBOND)
Denominator Category 27_Osteomyelitis
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for osteomyelitis not present on admission (Table 3; FTR27OSTEOMYD)
Denominator Category 28_Disseminated Intravascular Coagulation (DIC)
Denominator-eligible discharges with a secondary ICD-10-CM diagnosis code of disseminated intravascular coagulation (DIC) not present on admission (Table 3; FTR28DICD)
Denominator Category 29_Pyelonephritis
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for pyelonephritis not present on admission (Table 3; FTR29PYNEPHD)
Denominator Category 30_Postprocedural/Transfusion Complication
Denominator-eligible discharges with any listed secondary ICD-10-CM diagnosis code for postsurgical complication not present on admission (Table 3; FTR30POSTSURGD)
This measure uses submitted claims data to calculate the measure score. All data elements necessary to calculate this denominator are defined with the attached technical specifications.
Exclusions
DENOMINATOR OVERALL EXCLUSIONS (FOR ALL CATEGORIES)
Exclude discharges:
-Patients aged >90 years
-Admitted from a hospice facility (ADMSOUR = F)
-Do not resuscitate (DNR) status (ICD-10-CM Z66) present on admission (POA)
-Contradictory death information (reported date of death before admit date, death date before discharge date when patient was reportedly discharged alive, discharge disposition reported as died but enrollee has subsequent claims)
-No qualifying "operating room" procedure (Table 1) with a reported date
-First or only qualifying "operating room" procedure (Table 1) was outside appropriate time window for that claim (i.e., 4 or more days before the date of admission, or after the date of discharge)
-With an ungroupable MS-DRG (DRG=999)
-With missing discharge disposition (STUS_CD=missing), gender (SEX=missing), age (AGE=missing), quarter (DQTR=missing), year (YEAR=missing), or principal diagnosis (DGNS_CD1=missing)
-Discharged against medical advice (DISP=7)
Denominator Exclusions Category 1_Cardiac Event
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for cardiac event (Table 3; FTR1CARDEVENTD)
Denominator Exclusions Category 2_Congestive Heart Failure
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for congestive heart failure (Table 3; FTR2CHFD)
Denominator Exclusions Category 3_Hypotension/Shock/Hypovolemia
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for hypotension, shock, or hypovolemia (Table 3; FTR3SHOCKD)
-with any listed principal ICD-10-CM diagnosis code for trauma (Table 7)
Denominator Exclusions Category 4_Pulmonary Embolus/Deep Vein Thrombosis/Phlebitis
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for pulmonary embolus, deep vein thrombosis or phlebitis (Table 3; FTR5PEDVTPHD)
Denominator Exclusions Category 5_Cerebrovascular Accident (CVA)/TIA
-Exclude discharges: with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for stroke, cerebrovascular accident or transient ischemic attack (Table 3; FTR5CVAD)
Denominator Exclusions Category 6_Coma
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for coma (Table 3; FTR6COMAD)
Denominator Exclusions Category 7_Seizure
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for seizure (Table 3; FTR7SEIZD)
Denominator Exclusions Category 8_Delirium/Psychosis
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for psychosis (Table 3; FTR8PSYCHD)
Denominator Exclusions Category 9_Nervous System Complications
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for nervous system complications (Table 3; FTR9NERVSYSD)
Denominator Exclusions Category 10_Pneumonia/Pneumonitis
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for pneumonia or pneumonitis (Table 3; FTR10PNEUMOD)
Denominator Exclusions Category 11_Pneumothorax
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for pneumothorax (Table 3; FTR11PTXD)
Denominator Exclusions Category 12_Respiratory Compromise/Bronchospasm
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for respiratory compromise or bronchospasm (Table 3; FTR12RESPCOMPD)
Denominator Exclusions Category 13_Internal Organ Damage/Perforation
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for internal organ damage or perforation (Table 3; FTR13ORGDAMD)
Denominator Exclusions Category 14_Peritonitis
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for peritonitis (Table 3; FTR14PERITD)
Denominator Exclusions Category 15_GI Bleed and Blood Loss
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for GI bleeding or blood loss (Table 3; FTR15GIBLEEDD)
-with any listed principal ICD-10-CM diagnosis code for trauma (Table 7)
Denominator Exclusions Category 16_Sepsis
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for sepsis (Table 3; FTR16SEPSISD)
Denominator Exclusions Category 17_Deep Wound Infection/Wound Complication
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for deep wound infection or wound complication (Table 3; FTR17WOUNDD)
Denominator Exclusions Category 18_Renal Dysfunction
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for renal dysfunction (Table 3; FTR18RENALD)
Denominator Exclusions Category 19_Gangrene/Amputation
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for gangrene or amputation (Table 3; FTR19GANGAMPD)
Denominator Exclusions Category 20_Intestinal Obstruction/Ischemia
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for intestinal obstruction or ischemia (Table 3; FTR20INTOBSTISCHD)
Denominator Exclusions Category 21_Foreign Body
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for foreign body (Table 3; FTR21FORBODYD)
Denominator Exclusions Category 22_Pressure Injury
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for pressure injury (Table 3; FTR22PID)
Denominator Exclusions Category 23_Orthopedic Complication
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for orthopedic complication (Table 3; FTR23ORTHOCOMPD)
Denominator Exclusions Category 24_Hepatitis/Jaundice
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for hepatitis or jaundice (Table 3; FTR24HEPATD)
Denominator Exclusions Category 25_Pancreatitis
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for pancreatitis (Table 3; FTR25PANCD)
Denominator Exclusions Category 26_Necrosis of the Bone (Thermal or Aseptic)
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for necrosis of the bone (thermal or aseptic) (Table 3; FTR26NECBOND)
Denominator Exclusions Category 27_Osteomyelitis
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for osteomyelitis (Table 3; FTR27OSTEOMYD)
Denominator Exclusions Category 28_Disseminated Intravascular Coagulation (DIC)
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for disseminated intravascular coagulation (DIC) (Table 3; FTR28DICD)
Denominator Exclusions Category 29_Pyelonephritis
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for pyelonephritis (Table 3; FTR29PYNEPHD)
Denominator Exclusions Category 30_ Postprocedural/Transfusion Complication
Exclude discharges:
-with any listed principal ICD-10-CM diagnosis code (or secondary diagnosis present on admission) for postsurgical complication (Table 3; FTR30POSTSURGD)
This measure uses submitted claims data to calculate the measure score. All data elements necessary to identify denominator complications are defined with the attached technical specifications.
This measure uses submitted claims data, in combination with validated death data from the Medicare Beneficiary Summary File (or equivalent resources, such as a Vital Status File) to calculate the measure score. All data elements necessary to calculate these denominator exclusions are defined with the attached technical specifications.
Measure Calculation
See attached.
Not applicable; this measure is not stratified.
There is no minimum sample size to calculate the measure. If a hospital has fewer than 25 denominator eligible records, the hospital’s mortality rate and interval estimates will not be publicly reported. This approach is consistent with other CMS 30-day mortality measures used in public reporting programs.
Measure Record
Point of Contact
N/A
Donta Henson
7500 Security Boulevard
Baltimore, MD 21244
United States
Brittany Colip
Mathematica
600 Alexander Park Ste 100
Princeton, NJ 08540
United States
Importance
Evidence
The concept of “failure to rescue” (FTR) was originally developed by Jeffrey Silber and colleagues and adapted by Jack Needleman and colleagues. Over the past three decades, numerous studies have identified associations with multiple hospital characteristics and processes of care and rates of failure to rescue. The current measure is an updated and completely re-tested version of a previously CBE-endorsed measure #0353, “Failure to Rescue 30-day Mortality.” This measure was stewarded by Silber and colleagues at the Children’s Hospital of Philadelphia (CHOP) and used extensively by the research and quality improvement communities. CHOP allowed CBE endorsement to lapse in 2021.
Hospital Characteristics and Staffing
A series of seminal papers by Silber et al. and Needleman et al. established the relationship between several hospital characteristics and failure to rescue rates. Silber et al. (1992) examined 5,972 Medicare patients admitted for cholecystectomy and transurethral prostatectomy and found that failure to rescue was independent of severity of illness at admission, but was significantly associated with the presence of surgical housestaff and a lower percentage of board-certified anesthesiologists. The adverse occurrence rate was independent of these hospital characteristics. In a larger sample of 74,647 patients who underwent general surgical procedures in 1991-92, Silber et al. (1997) found lower failure to rescue rates at hospitals with high ratios of registered nurses to beds. Failure rates were strongly associated with risk adjusted mortality rates, as expected, but not with complication rates. Finally, among 16,673 patients admitted for coronary artery bypass surgery, failure to rescue rates were lower (whereas complication rates were higher) at hospitals with magnetic resonance imaging facilities, bone marrow transplantation units, or approved residency training programs (Silber et al., 1995). In a 2002 publication, Needleman and Buerhaus confirmed that higher registered nurse staffing (RN hours/adjusted patient day) and better nursing skill mix (RN hours/licensed nurse hours) were consistently associated with lower failure to rescue rates among major surgery patients from 799 hospitals in 11 states in 1997, even using administrative data to define complications. An increase from the 25th to the 75th percentile on these two measures of staffing was associated with 5.9% (95% CI, 1.5% to 10.2%) and 3.9% (95% CI, -1.1% to 8.8%) decreases, respectively, in the rate of failure-to-rescue among major surgery patients.
Other more recent individual studies have reported similar significant associations between failure to rescue and hospital characteristics, including nurse staffing levels (Aiken et al., 2011; Brooks Carthon et al., 2012; Ma et al., 2015; Silber et al., 2007), greater nurse education or advanced nurse skill mix (Kendall-Gallagher et al., 2011; Kutney-Lee et al., 2013; Silber et al., 2007), hospital volume (Gonzalez et al., 2014; Silber et al., 2009), nursing (ANCC) magnet status (Kutney-Lee et al., 2015; McHugh et al., 2013), resident-to-bed ratio or teaching status (Silber et al., 2007, 2009).
Several systematic reviews have reported confirmatory findings. A 2015 systematic review by Johnston et al. including 42 studies (some of which are described previously) identified several hospital characteristics associated with delayed escalation of care and higher FTR rates, including lower hospital volume, lower nurse staffing, and non-teaching status. The review identified 3 studies that found that mortality rates increased in patients with delayed escalation of care (odds ratio ranging from 2.1 to 3.1) and one study reporting that delayed transfer to the intensive care unit (ICU) was associated with 20% higher mortality compared to rapid transfer. A systematic review by Bourgon Labelle (2019) identified 15 studies finding significant associations between nurse staffing levels and improved failure to rescue rates (both in-hospital and 30-day) among patients with postoperative cardiac events. The review also identified 6 studies finding that a higher proportion of nurses with baccalaureate degrees was also significantly associated with lower 30-day failure to rescue rates. A systematic review by Twigg et al. (2019) identified nine studies reporting significant associations between nursing skill mix and failure to rescue rates among adult patients in acute care settings. In a systematic review by Audet et al. (2018), six studies were identified that reported significant associations between nursing education and lower risk of failure to rescue. Twigg and colleagues also found that the association between nursing education and failure to rescue was stronger for surgical patients than for non-surgical patients. In a meta-analysis of three studies, Liao et al. (2016) concluded that a 10% increase in nurses with a bachelor's degree or above was associated with a 5% reduction in risk of failure to rescue (OR: 0.95; 95% CI, 0.94-0.97; p<0.001).
References
- Aiken LH, Cimiotti JP, Sloane DM, Smith HL, Flynn L, Neff DF. Effects of nurse staffing and nurse education on patient deaths in hospitals with different nurse work environments. Med Care. 2011;49(12):1047-1053.
- Audet LA, Bourgault P, Rochefort CM. Associations between nurse education and experience and the risk of mortality and adverse events in acute care hospitals: A systematic review of observational studies. Int J Nurs Stud. 2018;80:128-146.
- Bourgon Labelle J, Audet LA, Farand P, Rochefort CM. Are hospital nurse staffing practices associated with postoperative cardiac events and death? A systematic review. PLoS One. 2019;14(10):e0223979.
- Brooks Carthon JM, Kutney-Lee A, Jarrín O, Sloane D, Aiken LH. Nurse staffing and postsurgical outcomes in black adults. J Am Geriatr Soc. 2012;60(6):1078-1084.
- Gonzalez AA, Dimick JB, Birkmeyer JD, Ghaferi AA. Understanding the volume-outcome effect in cardiovascular surgery: the role of failure to rescue. JAMA Surg. 2014;149(2):119-123.
- Johnston MJ, Arora S, King D, et al. A systematic review to identify the factors that affect failure to rescue and escalation of care in surgery. Surgery. 2015;157(4):752-763
- Liao LM, Sun XY, Yu H, Li JW. The association of nurse educational preparation and patient outcomes: Systematic review and meta-analysis. Nurse Educ Today. 2016;42:9-16.
- Kendall-Gallagher D, Aiken LH, Sloane DM, Cimiotti JP. Nurse specialty certification, inpatient mortality, and failure to rescue. J Nurs Scholarsh. 2011;43(2):188-194.
- Kutney-Lee A, Sloane DM, Aiken LH. An increase in the number of nurses with baccalaureate degrees is linked to lower rates of postsurgery mortality. Health Aff (Millwood). 2013;32(3):579-586.
- Kutney-Lee A, Stimpfel AW, Sloane DM, Cimiotti JP, Quinn LW, Aiken LH. Changes in patient and nurse outcomes associated with magnet hospital recognition. Med Care. 2015;53(6):550-557.
- Ma C, McHugh MD, Aiken LH. Organization of Hospital Nursing and 30-Day Readmissions in Medicare Patients Undergoing Surgery. Med Care. 2015;53(1):65-70.
- McHugh MD, Kelly LA, Smith HL, Wu ES, Vanak JM, Aiken LH. Lower mortality in magnet hospitals. Med Care. 2013;51(5):382-388.
- Needleman J, Berghaus P, Mattke S, Stewart M, Zelevinsky K. Nurse-staffing levels and the quality of care in hospitals. N Engl J Med. 2002;346(22):1715-1722.
- Silber JH, Williams SV, Krakauer H, Schwartz JS. Hospital and patient characteristics associated with death after surgery. A study of adverse occurrence and failure to rescue. Med Care. 1992;30(7):615-29.
- 15. Silber JH, Rosenbaum PR, Schwartz JS, Ross RN, Williams SV. Evaluation of the complication rate as a measure of quality of care in coronary artery bypass graft surgery. JAMA. 1995;274(4):317-323.https://pubmed.ncbi.nlm.nih.gov/7609261/
- 16. Silber JH, Rosenbaum PR, Williams SV, Ross RN, Schwartz JS. The relationship between choice of outcome measure and hospital rank in general surgical procedures: implications for quality assessment. Int J Qual Health Care. 1997;9(3):193-200. https://pubmed.ncbi.nlm.nih.gov/9209916/
- Silber JH, Romano PS, Rosen AK, Wang Y, Even-Shoshan O, Volpp KG. Failure-to-rescue: comparing definitions to measure quality of care. Med Care. 2007;45(10):918-925.
- Silber JH, Rosenbaum PR, Romano PS, et al. Hospital teaching intensity, patient race, and surgical outcomes. Arch Surg. 2009;144(2):113-121.
- Twigg DE, Kutzer Y, Jacob E, Seaman K. A quantitative systematic review of the association between nurse skill mix and nursing-sensitive patient outcomes in the acute care setting. J Adv Nurs. 2019;75(12):3404-3423.
Processes of Care
Studies also show that other processes of care can influence failure to rescue rates. Failure to rescue has been found to be associated with measures of a hospital’s aggressiveness of care (defined as the level of resources or inpatient spending), with hospitals that treat patients more aggressively having better surgical mortality and failure to rescue rates (Kaestner, 2010; Silber, 2010). Three recent systematic reviews have examined the relationship between the use of various hospital-based interventions and the risk of failure to rescue.
A 2022 systematic review by Burke et. al. including 52 articles identified three critical stages that lead to failure to rescue – failure to recognize complications, failure to relay information regarding complications, and failure to react in a timely and appropriate manner – and six types of interventions that can improve failure to rescue rates within healthcare organizations:
1. Staffing levels and education: Based on 14 studies (meta-analysis, retrospective cohort studies, cross-section studies, case-control studies, case reports, and a descriptive project), the authors found that FTR is highly sensitive to nursing care, specifically nurse-patient ratios, patient turnover and nurse staffing in non-ICU settings, staffing patterns, training and opportunity for simulation. For example, a cohort study demonstrated that after implementation of minimum nurse staffing levels in California, FTR rates decreased significantly more in California than in comparison states, with improvements of up to 32.9% (P < 0.05) in the final implementation period, across quartiles of baseline nurse staffing.
2. Detection, early warning signs (EWS) systems and checklists: Based on 8 studies (RCT, observational studies systematic review, cross-sectional studies and respect to follow up study), the authors observed the importance of early warning symptom detection protocols and timely and appropriate escalation. For example, a randomized controlled trial (RCT) demonstrated improved patient management (SWAT-M, P < 0·001) and nontechnical skills (P = 0·043) between baseline and final ward rounds, whereas the control group showed no improvement (P = 0·571 and P = 0·809, respectively). A small learning effect was seen with improvement in patient assessment (SWAT-A) in both groups (P < 0·001).
3. Surveillance, communication and electronic monitoring: Based on 6 studies (retrospective cohort study, cross-sectional study, observation of a pilot, perspective single blinds observational study and a retrospective observational study of her control), the authors underscore the importance of nursing communication and continuous monitoring. For example, a retrospective cohort study demonstrated that when nursing surveillance was performed at least 12 times a day, there was a significant (P = 0.0058) decrease in the odds of experiencing failure to rescue (OR = 0.52) compared with when surveillance was delivered an average of <12 times a day.
4. Medical emergency and rapid response teams (RRT): Based on 8 studies (cluster RCT, retrospective audit, cross-sectional survey, case control, retrospective observational, descriptive/competitive study, longitudinal study and interrupted times serious population base study), the authors observe that significant variation in the design and reporting of studies examining medical emergency teams (METs) and RRTs limits the ability to draw clear conclusions regarding effectiveness. For example, a cluster RCT demonstrated similar incidence of the composite primary outcome in the control and MET hospitals (5.86 versus 5.31 per 1000 admissions, P = 0.640), as well as of the individual secondary outcomes (cardiac arrests, 1.64 versus 1.31, P = 0.736; unplanned ICU admissions, 4.68 versus 4.19, P = 0.599; and unexpected deaths, 1.18 versus 1.06, P = 0.752). A reduction in the rate of cardiac arrests (P = 0.003) and unexpected deaths (P = 0.01) was seen from baseline to the study period for both groups combined, suggesting an effect of study participation unrelated to the MET program.
5. Relaying information about complications: Based on 9 studies (cohort study, literature review of six studies, cross-sectional survey, multi center qualitative study, observational, perspective observational and observational, questionnaire-based), the authors conclude that interprofessional communication and nurse physician relationship are of paramount importance, and recommended use of SBAR as a communication tool. For example, one study involved prospective collection of predefined surgical critical events and communications, patient interviews, and sporadic clinical questioning of junior clinicians. The authors reported that of 80 critical patient events identified across four hospitals, 26 (33%) were not communicated to attending surgeons. Although residents felt that attending contact was unnecessary for safe patient care in 61 (76%) of these events, discussions with attending physicians changed management in 33% (18/54) of cases in which they occurred.
6. Reacting to a patient in a timely manner with the correct evidence-based management: Based on 3 studies (audit of single center two units, retrospective cohort with contemporaneous control group, and retrospective cohort), the authors found that timely and evidence-based interventions have a significant impact on patient outcomes; for example, timely administration of antibiotics to patients with sepsis.
A 2015 systematic review by Johnston et al. identified several interventions that can improve timely escalation of care, including new vital sign charts and improved documentation, escalation protocols, and communication tools. Four studies found that these interventions increased the number of escalation-of-care calls or physician communications regarding deterioration. One pre-post cohort study found that an escalation protocol led to a non-significant decrease in in-hospital cardiac arrests (3% vs. 9% pre-implementation) and a significant decrease in ICU admission rates (23% vs. 46% pre-implementation, p<0.001). A second pre-post cohort study found that use of a new vital signs chart led to a non-significant decrease in in-hospital cardiac arrest (0.5% vs. 1.8% pre-implementation) and a significant decrease in mortality (0.6% vs. 2.6% pre-implementation).
In the recent Making Healthcare Safer III report, Hall et al. (2020a) examined two patient safety practices with the potential to impact failure to rescue rates – patient monitoring systems and rapid response teams. Of the 8 included studies examining the impact of patient monitoring systems, there was moderate but inconsistent evidence that systems with continuous monitoring lead to reductions in failure to rescue events. Hall et al. (2020a, 2020b) identified 10 studies (including 3 meta-analyses and 3 systematic reviews) examining the impact of rapid response teams (RRTs) on failure to rescue events. This systematic review found that the implementation of RRTs was associated with decreases in inpatient mortality and in-hospital cardiac arrest. Two of the three meta-analyses found that RRT implementation significantly decreased mortality rates among adult inpatients (pooled relative risk [RR] range, 0.87-0.88), while the third found no difference in overall mortality (pooled RR, 0.92; 95% CI, 0.82-1.04). Three meta-analyses identified overall decreases in non-ICU cardiac arrest after RRT implementation (pooled RR range, 0.62-0.65). Hall et al. reported mixed results on the impact of RRT on ICU transfer rates – one meta-analysis including 10 studies found no association while one systematic review found that RRTs reduced unplanned ICU admissions.
References
- Burke JR, Downey C, Almoudaris AM. Failure to Rescue Deteriorating Patients: A Systematic Review of Root Causes and Improvement Strategies. J Patient Saf. 2022;18(1):e140-e155.
- Hall KK, Lim A, Gale B. Failure To Rescue. In: Hall KK, Shoemaker-Hunt S, Hoffman L, et al. Making Healthcare Safer III: A Critical Analysis of Existing and Emerging Patient Safety Practices [Internet]. Rockville (MD): Agency for Healthcare Research and Quality (US); 2020a. Available from: https://www.ncbi.nlm.nih.gov/books/NBK555513/
- Hall KK, Lim A, Gale B. The Use of Rapid Response Teams to Reduce Failure to Rescue Events: A Systematic Review. J Patient Saf. 2020b;16(3S Suppl 1):S3-S7.
- Johnston MJ, Arora S, King D, et al. A systematic review to identify the factors that affect failure to rescue and escalation of care in surgery. Surgery. 2015;157(4):752-763
- Kaestner R, Silber JH. Evidence on the efficacy of inpatient spending on Medicare patients. Milbank Q. 2010;88(4):560-594.
- Silber JH, Kaestner R, Even-Shoshan O, Wang Y, Bressler LJ. Aggressive treatment style and surgical outcomes. Health Serv Res. 2010;45(6 Pt 2):1872-1892.
Measure Impact
By using failure-to-rescue (FTR), a risk-standardized measure of death after an adverse occurrence, hospitals can identify opportunities to improve their quality of care. Hospitals and health care providers benefit from knowing not only their institution’s mortality rate, but also their institution’s ability to rescue patients after clinical deterioration. The measure is especially important if hospital resources needed for preventing complications are different from those needed for rescue. We anticipate that this measure will encourage hospitals to focus on early identification and rapid treatment of complications, thereby improving the overall quality of care. Failure to rescue measures have been repeatedly validated by their consistent association with nurse staffing, nursing skill mix, technological resources, rapid response systems, and other activities that improve early identification and prompt intervention when complications arise after surgery.
Performance Results from Beta Testing:
Risk-standardized rates show substantial variation in performance scores across the 2,055 eligible facilities with at least 25 qualifying denominator records. Specifically, the distribution of 30-day Failure to Rescue risk-standardized death rates in our test data is as follows:
5th percentile: 0
15th percentile: 21.15
25th percentile: 29.33
35th percentile: 35.15
45th percentile: 40.35
55th percentile: 46.88
65th percentile: 53.35
75th percentile: 60.95
85th percentile: 71.64
95th percentile: 98.01
Median: 43.48
Mean: 46.62
This empirical analysis demonstrates considerable opportunity for improvement if facilities at the 75th percentile (60.95 risk-standardized deaths per 1,000 qualifying surgical cases) could move across the interquartile range to the 25th percentile (29.33 risk-standardized deaths per 1,000 surgical cases), which would represent a 50% decrease in the frequency of deaths after postoperative complications.
See Table 1 logic model attachment for a distribution of performance scores for the current CMS PSI 04 compared to the proposed measure. Compared with the current CMS PSI 04 measure that is used for public reporting, the proposed measure has a much higher minimum volume threshold (25 versus 3), covers over 8 times more denominator patients, and captures about 2.1 times more numerator events (deaths). The numerator increase is largely due to the application of this measure to both Medicare Advantage and FFS enrollees, as well as the inclusion of deaths after hospital discharge but within 30 days of the index operative procedure.
This measure has been designed and tested to replace CMS PSI 04, which is currently being used in the Hospital Inpatient Quality Reporting (HIQR) Program (00134-02-C-HIQR, formerly CBE #0351). This redesign is intended to address stakeholder concerns about the existing PSI 04 measure, which include:
1.Complications sometimes develop BEFORE the index operation in PSI 04, even before transfer to the index hospital (I.e., the operation is part of an effort to “rescue” the patient).
2. The heterogenous cohort includes patients with very high-risk surgery (e.g., trauma surgery, burn surgery, organ transplants, intracranial hemorrhage) and very low-risk surgery (e.g., eye, ear, urolithiasis).
3. Mean length of stay and prevalence of early discharge to post-acute facilities vary across hospitals, causing bias in comparing performance.
4. PSI 04 appears to slightly disadvantage major referral centers, even after risk-adjustment.
The respecified FTR measure will create a more homogenous denominator population and capture post-discharge deaths within 30-days after the first denominator-qualifying operation. This redesign is intended to better align the measure with the previously CBE-endorsed measure of Failure-to-Rescue: 30-Day Mortality (CBE #0353, endorsed 2008, renewed 2012 and 2015, allowed to lapse 2021).
Measures of failure-to-rescue among hospitalized surgical patients have been found to be useful by multiple stakeholders in the United States. For example, in the Fiscal Year (FY) 2022 Medicare Hospital Inpatient Prospective Payment System (IPPS) Proposed Rule (CMS-1752-P, April 2021), CMS proposed to retire PSI 04 from use in CMS programs. In response, CMS received many communications from patients, caregivers, patient advocacy organizations, employers and employer coalitions, and others. These communications clearly articulated the perceived value of CMS PSI 04 as a broad measure of postoperative mortality and hospitals’ skill at rescuing patients who experience complications. In response, CMS did not finalize the proposal to retire PSI 04 and invested in improving it in response to stakeholder feedback.
Equity
Equity
Using data from all 2,907 hospitals in our test data set, we conducted a social disparities analysis and found:
-Hispanic patients have similar risk of Failure to Rescue (OR=0.93; 95% CI, 0.82-1.05) as non-Hispanic patients, after adjusting for age and other factors in the risk-adjustment model.
-Black patients (OR=0.96; 95% CI, 0.91-1.01) and patients of "other" race (OR=1.06; 95% CI, 0.94-1.20) have similar risk of Failure to Rescue as White patients, after adjusting for age and other factors in the risk-adjustment model.
-Risk of Failure to Rescue is unrelated to sex, after adjusting for age and other factors in the risk-adjustment model. Sex was considered as a risk-adjustment feature but was found to have no marginal predictive value.
-Analyses of observed, expected, and risk-adjusted rates in all of the above patient cohorts confirm that the comorbidities, operative, and demographic factors in the risk-adjustment model account for some increased risk of Failure to Rescue among Black patients and patients of “other” race (average expected rate 5.81% and 5.71%, respectively, versus 4.18% among White patients), and that any residual bias is neither clinically nor statistically significant.
Empirical analyses confirm that this measure is neutral across social risk groups, including race, ethnicity, urbanicity/rurality, and sex, due to adjustment for all the patient characteristics described above. Age was explicitly included in the risk-adjustment model, so its effect was directly removed. These findings are as expected based on our conceptual model.
Feasibility
Feasibility
Because this measure is based on readily available administrative claims data, feasibility is not an issue. A similarly designed measure (CMS PSI 04) has been used by CMS for over a decade. No difficulties have been reported with respect to data collection, availability of data, missing data, timing and frequency of data collection, sampling, patient confidentiality, or time and cost of data collection. Hospitals routinely generate and transmit claims in a timely manner for all Medicare beneficiaries.
No feasibility assessment was completed due to the reasons outlined above.
Proprietary Information
There are no fees associated with use of this claims-based measure. The measure specifications will be available upon request through the CMS QualityNet Help Desk.
Scientific Acceptability
Testing Data
This measure was originally developed using data on Medicare FFS discharges from Inpatient Prospective Payment System (IPPS) hospitals, including hospitals in Maryland and excluding Veterans Administration hospitals, from for the period July 1, 2019, through December 31, 2019 and July 1, 2020 through June 30, 2021. Q1 and Q2 2020 data were excluded due to the blanket Extraordinary Circumstance Exception (ECE) for COVID-19. These data included roughly 12.4 million inpatient discharges from 3,357 hospitals where Medicare was the primary payer.
The measure was then tested on Medicare data from January 1, 2021 through June 30, 2022, including monthly inpatient claims files (Research Identifiable Files, or RIF) and Medicare Beneficiary Summary Files, These data included roughly 10.5 million discharges from 3,163 hospitals where Medicare was the primary payer. We used CMS+VA PSI v13.0 software to calculate the number of cases meeting the definition for the numerator and denominator for the current CMS PSI 04 measure and the proposed failure-to-rescue measure. We specifically evaluated the impact of changing the numerator definition from in-hospital death to 30-day death, with the 30-day window starting on the day of the first “operating room” procedure.
Not applicable.
Descriptive characteristics of the hospitals and Medicare FFS population included in testing are shown in Tables 2 and 3 of the logic model attachment.
The test data set includes 417,054 inpatient encounters from 2,163 Medicare Inpatient Prospective Payment System hospitals, including Maryland hospitals. Of these hospitals, 2,055 met the minimum denominator threshold of 25 for reporting their Failure to Rescue rate. These hospitals are very diverse, representing all bed size categories, teaching status categories, nursing skill mix and staffing categories, and location (urban/rural) categories. Test hospitals are situated in all 50 US states and the District of Columbia.
Reliability
We applied split-half and test-retest approaches to estimate the reliability of this risk-adjusted measure at the accountable entity (hospital) level, using the intracluster correlation coefficient (ICC) as an estimator. As formulas are not allowed in the online form, see logic model attachment pg. 9-10 for the methodology.
By design, hospital-level risk-adjusted outcome measures are centered around a global mean with an approximately normal distribution (allowing for the fact that the tails of the distribution may be augmented with hospitals that are true quality outliers). Because this ICC depends only on the ratio of between-hospital to within-hospital estimated variance components, and the relevant denominator for each hospital, we can estimate reliability as a function of the hospital’s denominator size, using an application of the Spearman-Brown prophecy formula. We applied this methodology to hospital subsamples that were formed by randomly dividing the available year of patient data from each hospital into two, then executing the measure code separately on each split-half, to yield two estimates per hospital.
The higher the ICC, the greater the statistical reliability of the measure, and the greater the proportion of variation that can be attributed to systematic differences in performance across hospitals (i.e., signal as opposed to noise). We used the rubric established by Landis and Koch (1977) to interpret ICCs:
0 – 0.2: slight agreement
0.21 – 0.39: fair agreement
0.4 – 0.59: moderate agreement
0.6 – 0.79: substantial agreement
0.8 – 0.99: almost perfect agreement
1: perfect agreement
References
- Dickens, William T. "Error components in grouped data: is it ever worth weighting?." The Review of Economics and Statistics (1990): 328-333.
- Landis, J. Richard, and Gary G. Koch. "The measurement of observer agreement for categorical data." biometrics (1977): 159-174.
- Spearman-Brown Prophecy Formula” in: Frey, B. (2018). The SAGE encyclopedia of educational research, measurement, and evaluation (Vols. 1-4). Thousand Oaks, CA: SAGE Publications, Inc. doi: 10.4135/9781506326139
Signal-to-noise reliability was estimated as an intraclass correlation coefficient based on a two-way mixed model with facility random effects (C,1), inflating the denominator and numerator for each hospital from 18 months to a full 24-month reporting period. To further improve reliability for public reporting, we recommend empirical Bayesian shrinkage (i.e., smoothing) to reduce random noise, in accord with standard methods for all AHRQ and CMS Patient Safety Indicators. The smoothed rate is a weighted average of the reference population rate and the local (hospital) risk-adjusted rate. If the data from the individual hospital include many observations and provide a numerically stable estimate of the rate, then the smoothed rate will be very close to the risk-adjusted rate, and it will not be heavily influenced by the reference population rate. Conversely, the smoothed rate will be closer to the reference population rate if the hospital rate is based on a small number of observations and may not be numerically stable, especially from year to year. As a weighted average of the risk-adjusted rate and the rate observed in the reference population, the smoothed rate is calculated with a shrinkage estimator, as described in this report: https://qualityindicators.ahrq.gov/Downloads/Resources/Publications/2023/Empirical_Methods_2023.pdf .
Signal-to-noise reliability was estimated as an intraclass correlation coefficient based on a two-way mixed model with facility random effects (C,1).
- Minimum: 0.231
- 25th percentile: 0.388
- Median: 0.568
- 75th percentile: 0.738
- Maximum: 0.973
Please note the functionality of the decile table below was not working at the time of submission. As such, the decile information is included below for reference in the following format: Decile #/Reliability value/ # Entities/Total Persons
Overall/0.7039(mean)/2055/1087624
Minimum/0.2314/21/525
Decile 1/0.2571/205/15853
Decile 2/0.3248/206/21776
Decile 3/0.3879/206/29419
Decile 4/0.4671/206/40024
Decile 5/0.5379/205/53027
Decile 6/0.6016/205/69384
Decile 7/0.6697/205/92901
Decile 8/0.7384/206/129893
Decile 9/0.8106/205/199744
Decile 10/0.8861/206/435603
Maximum/0.9733/1/8099
Failure to Rescue demonstrates moderate signal-to-noise reliability at most test facilities, based on a 24-month reporting period with both Medicare FFS and Medicare Advantage enrollees, as the mean and median ICC values equal 0.704 and 0.568, respectively. The comparable metrics for the currently reported version of CMS PSI 04, which is limited to Medicare fee-for-service patients, are 0.256 and 0.209, respectively, based on CMS+VA PSI v13 software applied to the 2023-reported performance period. The percentage of all eligible entities with reliability of at least 0.4 for Failure to Rescue is approximately 73% (based on a 24-month reporting period), versus 25% for the currently reported version of CMS PSI 04.
As with any 30-day mortality measure, reliability at the hospital level varies in accord with the size of the hospital and its eligible denominator. Minimum volume thresholds can be applied and adjusted, as needed, to address low reliability at low-volume hospitals. By regulation, the current minimum (denominator) volume threshold for all CMS 30-day risk-standardized mortality measures is 25. Overall, testing results showed that Failure to Rescue, as currently specified, can distinguish true performance across hospitals of typical size and volume.
| | Overall | Minimum | Decile_1 | Decile_2 | Decile_3 | Decile_4 | Decile_5 | Decile_6 | Decile_7 | Decile_8 | Decile_9 | Decile_10 | Maximum |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Reliability | 0.7039 (mean) | 0.2314 | 0.2571 | 0.3248 | 0.3879 | 0.4671 | 0.5379 | 0.6016 | 0.6697 | 0.7384 | 0.8106 | 0.8861 | 0.9733 |
| Mean Performance Score | 2055 | 21 | 205 | 206 | 206 | 206 | 205 | 205 | 205 | 206 | 205 | 206 | 1 |
| N of Entities | 1087624 | 525 | 15853 | 21776 | 29419 | 40024 | 53027 | 69384 | 92901 | 129893 | 199744 | 435603 | 8099 |
Validity
Convergent validity refers to the degree to which multiple measures of a single underlying concept are positively correlated with each other. To assess the convergent validity of the measure, we have compared the measure results with related measures of patient safety and outcomes. For this comparison, we drew on hospital-level quality measure results publicly available on data.Medicare.gov. Using Spearman rank correlation coefficients, we compared hospital-level failure-to-rescue rates with rates of risk-standardized 30-day readmission and mortality rates (e.g., hospital-wide unplanned all-cause readmissions), complications for hip/knee replacement patients and a composite measure of patient safety and adverse events. Correlations among these measures would support the validity of the failure-to-rescue measure because they measure a similar quality construct of patient safety. However, we do not expect strong correlations because patient safety is a complex construct, and these measures differ from the failure-to-rescue measure in terms of the populations and conditions being measured.
Known groups validity is a type of construct validity that focuses on a measure’s ability to discriminate between groups of measured entities that are known to differ on the underlying latent construct. With respect to hospital quality and safety, prior research has demonstrated several “known groups” that can be identified from the available data:
-Hospital resident-to-bed ratio, stratified as major teaching/academic (at least 0.25 fulltime equivalent [FTE] residents per bed), minor teaching/academic (more than 0 but less than 0.25 FTE residents per bed), and non-teaching
-Hospital nurse-to-bed ratio, stratified as highly staffed (more than 2.0 FTE licensed nurses per bed), moderately staffed (1.0-2.0 nurses per bed), poorly staffed (less than 1.0 nurses per bed)
-Hospital nurse skill mix, estimated as the proportion of all nursing FTEs or nursing hours that are provided by registered nurses (versus licensed vocational/practical nurses), stratified as relatively low (less than 85%), medium (85-97.5%), and high (over 97.5%)
-Hospital urban/rural location.
We hypothesized that failure-to-rescue rates would be lower at major teaching hospitals, urban hospitals, and hospitals with high nurse staffing and skill mix than at non-teaching hospitals, rural hospitals, and hospitals with low nurse staffing and skill mix, respectively.
Face validity refers to the degree to which evidence, clinical judgement, and theory support the interpretations of a measure score. Face validity is an assessment by experts that determines the extent to which a measure, at face value, appears to reflect what it is intended to assess. To determine face validity, we obtained input from members of the TEP to determine whether they think the measure as specified will help inform consumers and help providers improve quality.
Convergent validity was assessed using other measures of hospital quality that are used in Federal programs, focusing on measures that do not cover postoperative mortality. For all but one of these comparisons, the proposed measure demonstrates higher convergent validity than the current CMS PSI 04 measure (Table 4 in the logic model attachment). Of note, the Spearman rank correlation coefficient between this measure and the 30-day hospital-wide unplanned readmission measure was 0.229 (p<0.001). These findings show the expected direction and strength. Hospitals with higher nurse staffing and skill mix tend to have lower death rates after serious postoperative complications. Hospitals that identify complications late or fail to treat them aggressively tend to have higher 30-day readmission rates and higher death rates after serious postoperative complications.
As shown in Table 5 of the logic model attachment, the data support these hypotheses for all “known groups” except rural/urban location. Full-time equivalent nurse-to-bed ratio was classified as <1; 1-2; or 2. Relative to the 496 hospitals with the lowest nurse staffing, the 1,266 hospitals with intermediate nurse staffing had an overall rate ratio of 0.98, and the 445 hospitals with the highest nurse staffing had an overall rate ratio of 0.84 (p<0.001). Similar results were found for nursing skill mix; 872 hospitals with the highest ratios of RN-to-total nurse staffing had an overall rate ratio of 0.83 (p<0.001), compared with the 328 hospitals with the lowest ratios.
Face validity results are as follows:
- 9 of 10 members (90%) voted “yes” that the measured outcome (rate of 30-day mortality among surgical inpatients with complications) provides a representation of relevant quality in a facility.
- 9 of 10 members (90%) voted “yes” that implementation of the measure in hospital inpatient quality reporting programs (in place of current PSI 04) is likely to lead to improve quality of care by reducing the frequency of failure to rescue.
- 5 of 5 members (100%) who are employed by a “measured entity” (i.e., employed or affiliated with hospital organizations) voted “yes” that the proposed measure is easy to understand and may be useful for decision-making.
The one member who disagreed felt that the proposed denominator expansion (adding patients who experience less serious complications after surgery) makes the measure less relevant to identifying hospitals’ performance in rescuing higher risk/serious cases. The member indicated that other CMS mortality measures address lower risk cases, while PSI 04 is unique in its focus on patients with a very high risk of death. In response, the team highlighted that there is only one current mortality measure that focuses on surgical cases and that measure is limited to CABG. This proposed expansion is bringing a new and broader population of surgical patients into the measurement sphere. These patients better represent “typical” surgical patients undergoing bariatric surgery, orthopedic surgery, cancer surgery, colorectal surgery, etc. Only if patients with mild-to-moderate complications are brought into the denominator can we focus attention on preventing the progression of complications from mild to serious, which is the core of the failure-to-rescue concept. The improvements to this measure make it unique as a measure of surgical outcomes (failure-to-rescue) across a broad set of non-emergency procedures.
Systematic assessment of face validity of the performance measure score confirms that the score is believed to accurately reflect hospital performance with respect to postoperative care, and to distinguish good from poor performance. The only negative vote in the expert panel process was motivated by concern about modifying the denominator population, compared with the current CMS PSI 04 measure, by excluding certain high-risk patients such as multiple trauma, burns, and transplants, and refocusing the denominator population on general surgery, orthopedic surgery, and cardiovascular surgery. However, this change was motivated by over a decade of feedback from the user community and both public and private stakeholders.
Empirical testing results confirm that the proposed Failure to Rescue measure, which is designed to align with the prior CBE-endorsed measure #0353 (“Failure to Rescue 30-day Mortality”), has superior convergent validity and known groups validity compared with the measure currently used in CMS programs, CMS PSI 04. These properties are also consistent with the performance of #0353, as previously reported to the CBE.
Risk Adjustment
There are established risk factors for failure to rescue, many of which are outside hospitals’ control (e.g., age, comorbidity burden). Risk factors for failure to rescue can be categorized into three groups – (1) patient risk factors for mortality within 30 days of surgery, such as age, comorbidities, or preoperative ‘do not resuscitate’ orders; (2) social risk factors that can influence patient risk, such as patient functional status, race/ethnicity, or socioeconomic status, and; (3) hospital factors, such as nurse and resident staffing, staff skill mix, hospital volume and technological resources. Patient attributes (demographics, comorbid conditions, clinical signs and symptoms, functional risk factors, and others) present at the start of care are integral components of the risk model, in that they directly influence the measured outcome and hospitals have less control. Care processes and intermediate factors (or mediators) can influence failure to rescue rates. These factors are largely within a hospital’s control and are therefore not considered as risk factors. These process factors are summarized in the Importance section. Examples of models that have been included in published studies are included in Table 6 of the logic model attachment.
Because of the large number of measured entities (2,907 with at least one denominator record; 2,055 with at least 25 denominator records), we are unable to report descriptive statistics for the risk variables at the entity level. For additional details regarding the overall frequency of all risk factors (and risk factors that were considered but not selected for the final model), please refer to Table 3 in the logic model attachment. Mean age varies across measured entities from a minimum of 63.9 years to a maximum of 79.4 years, with 25th, 50th, and 75th percentile values of 72,0, 73.3, and 74.4 years, respectively. Mean values of the Elixhauser (AHRQ) Comorbidity Risk of Mortality Index vary across measured entities from a minimum of –5.5 to a maximum of 22.4, with 25th, 50th, and 75th percentile values of 4.0, 6.0, and 7.8, respectively. Finally, as a summary measure of variation, the expected rate of Failure to Rescue varies across measured entities from a minimum of 2.21 per 1,000 surgical cases to a maximum of 130.67 per 1,000 surgical cases, with 25th, 50th, and 75th percentile values of 34.79, 44.28, and 52.51, respectively.
The final risk-adjustment model was estimated using cluster-adjusted multivariable logistic regression to optimize calibration, after testing both logistic and probit link functions. The model was also estimated using a mixed-level logistic model with hospital random effects, but the results (including the confidence intervals surrounding parameter estimates) were virtually unchanged, compared with simpler form models. All risk factors were dichotomous (0/1) except for:
-age, which was tested in both piecewise linear and categorical forms;
-discharge quarter, which was tested as a set of dummy variables to capture secular trends in risk-standardized mortality over time (and unmeasured secular trends in case mix due to the post-pandemic backlog in elective surgery);
-Modified Diagnosis-Related Groups (MDRGs) representing aggregates of adjacent CMS MS-DRGs without comorbidities or complications, with comorbidities or complications, or with major comorbidities or complications, which were tested as a fully saturated set of dummy variables;
-AHRQ’s default Clinical Classifications Software Refined (CCSR) for International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM)-codes, applied to the principal diagnosis on each record, which were tested as a fully saturated set of dummy variables; and
-Elixhauser Index for Risk of In-hospital Mortality, which was tested as a continuous variable.
MDRGs were used to adjust for the type of operation for which the patient was admitted (excluding tracheostomy, which often follows a period of postoperative respiratory failure). CCSRs were used to adjust for the principal reason for the patient’s admission to the hospital. The Elixhauser Index was used to adjust for the combined effect of multiple comorbidities, including comorbidities that were not sufficiently frequent or sufficiently impactful to be selected as independent risk factors.
All data came from the fields available on Medicare FFS claims and Medicare Advantage shadow claims (inpatient encounter records), including ICD-10-CM diagnosis codes for comorbidities present on admission, ICD-10-CM principal diagnosis codes, ICD-10-PCS procedure codes affecting the CMS MS-DRG assignment, hospital-reported source of admission (i.e., transfer from another hospital), and demographic fields for age, sex, and discharge year and quarter. Interactions between COVID-19 present on admission and discharge quarter were used to account for the changing impact of COVID-19 over time, as population immunity has improved and more effective treatments have become available. Two transfer variables were created to adjust for the possibility that patients transferred from one hospital to another for an operation may be at higher risk than patients who remain at the hospital where they presented, even after adjusting for other measured patient characteristics. One of these features is based on transfers reported by the receiving hospital, and the other is based on transfers identified from Medicare claims data even without reporting by the receiving hospital.
Guided by the conceptual model, we developed the baseline risk adjustment model for FTR using the following process.
1. Randomly partitioned the full denominator data into an 80% training set and a 20% hold-out (model performance or evaluation) test set.
2. Created contingency tables for all categorical features to identify any that had zero cells for either the positive or negative outcome. These features were not considered further due to anticipated model convergence problems (i.e., quasi-complete separation). For continuous variables, such as age, we ran locally weighted bivariate regressions (i.e., locally weighted scatterplot smoothing, or LOWESS) to understand the functional form of the relationship. This analysis confirmed that the risk of FTR was not linearly related to age, except for the limited age range between 70 and 90 years.
3. Fit one model using the least absolute shrinkage and selection operator (LASSO) on the training set using 10-fold cross-validation (CV). This step helped to assess model fit on the training set, while facilitating parameter tuning (e.g., the lambda regularization parameter in the cross-validation [CV]-based LASSO). We chose the final model where the regularization parameter (lambda) was set to lambda1se, i.e., “one-standard-error” (i.e., the largest lambda at which the mean squared error (MSE) is within one standard error of the minimum MSE.). This rule is standard practice for improving generalization, and its suitability was confirmed using the hold-out test set.
4. Given that Lasso was able to provide a robust solution, with consistent selection of the same 120±5 features, we did not use other penalized regression approaches (e.g., Elastic Net).
5. The final risk-adjustment model was a cluster-adjusted logistic regression model. The model was estimated on the entire dataset using the set of features selected by Lasso through 10-fold cross-validation and testing on the hold-out test set.
6. The risk-adjustment model was also tested with additional social drivers of health variables (Medicaid insurance, Hispanic ethnicity, Race), considered individually and collectively.
References
- T. Hastie, R. Tibshirani, J. Friedman, The Elements of Statistical Learning (Springer, 2001), vol. 1.
We summarize model performance using the following measures:
-Overall model discrimination as assessed by C-statistic. The C-statistic is the area under the receiver-operator curve (i.e., AUC) that measures the discriminative ability of a regression model across all levels of risk. It also describes the probability that a randomly selected patient who experienced a fall with injury had a higher expected value than a randomly selected patient who did not experience that event. The AUC was 0.816 in the holdout test set (based on Lasso) and 0.818 for the final logistic model. These values indicate strong discrimination performance, relative to a random classifier with AUC=0.5.
-The precision-recall (PR) curve and the area under the curve (AUPRC). The PR curve and AUPRC are less sensitive to data imbalance or class imbalance (i.e., very rare events) than the AUC. The AUPRC was 0.184 in the holdout test set (based on Lasso), indicating good prediction at the individual patient level relative to a random classifier with AUPRC=0.043.
-Model calibration was assessed across deciles of patient risk using Hosmer-Lemeshow plots. The deciles of risk are ten mutually exclusive groups containing equal numbers of discharges, ranging from very low-risk patients (according to the model) to high-risk patients. We do not provide Hosmer-Lemeshow test statistics because, given the large sample size of our data, the null hypothesis is almost always rejected. Moreover, the plots provide more detail on model fit than the overall Hosmer-Lemeshow statistic. Because over 43% of events occurred in the highest-risk decile, and over 63% occurred in the highest-risk quintile, the decile analysis is statistically unstable. However, the analysis suggests overestimation of risk among low-risk patients in the bottom five deciles (i.e., observed-to-expected ratios of 0.64-0.84 among patients with death rates under 2%), but very accurate estimation among high-risk patients in the top five deciles (i.e., observed-to-expected ratios of 0.99-1.11 among patients with death rates over 2%). Alternative link functions are being tested to better account for the overestimation of risk among low-risk patients.
See above.
126
Use & Usability
Use
Usability
No facility is expected to have a zero rate for this measure because it targets rescue from severe conditions that have a non-zero death rate. In many cases, rescue procedures may be unsuccessful or the decision to discontinue them may be made. When treatment of a complication has been unsuccessful, providers and family members often decide to order “do not resuscitate” or “palliative care” or “comfort measures only”; these choices do not affect the measure because they are generally consequences of the patient’s clinical deterioration, not direct causes of it.
However, there are evidence-supported interventions that hospital can implement to improve timely identification of clinical deterioration and treatment of preventable complications, including improved nurse staffing, simulation training, standardized communication tools, electronic monitoring and/or warning systems, and rapid response systems.
Comments
Staff Preliminary Assessment
CBE #4125 Staff Assessment
Importance
Strengths:
- The developer provides a logic model depicting various structural changes and procedures that can be implemented by hospitals to improve the timely recognition of clinical deterioration and treatment, which will lead to reduced mortality associated with failure to rescue.
- The developer posits that with this measure, hospitals can identify opportunities to improve their quality of care and that this measure will encourage hospitals to focus on early identification and rapid treatment of complications, thereby improving the overall quality of care.
- The developer cites various studies that show various hospital characteristics, such as higher nurse-to-bed ratios, more advanced nurse skill mix, greater hospital volume, and others have been shown to reduce failure to rescue rates. In addition, use of technology-supported interventions (such as patient monitoring systems and rapid response teams), standardized communication tools, or simulation training can improve timely recognition and response to clinical deterioration and reduce failure to rescue.
- The developer states that this measure is a respecified version of CBE#0353 - Failure to Rescue 30-day Mortality, which is no longer endorsed. It also is intended to replace the CMS PSI 04, which is currently being used in the Hospital Inpatient Quality Reporting (HIQR) Program.
- The developer reports initial risk-standardized rates for the measure across 2,055 facilities with 25 qualifying records. The mean is 46.2 with an interquartile range of 29.33 to 60.95.
- The developer states that these communications clearly articulated the perceived value of CMS PSI 04 as a broad measure of postoperative mortality and hospitals’ skill at rescuing patients who experience complications.
Limitations:
- The developer did not provide direct patient input for this measure but does note the communications received from the patient community with respect to the retirement of the PSI 04.
Rationale:
- The developer provides a logic model depicting various structural changes and procedures that can be implemented by hospitals to improve the timely recognition of clinical deterioration and treatment, which will lead to reduced mortality associated with failure to rescue.
- The developer posits that with this measure, hospitals can identify opportunities to improve their quality of care and that this measure will encourage hospitals to focus on early identification and rapid treatment of complications, thereby improving the overall quality of care.
- The developer cites various studies that show various hospital characteristics, such as higher nurse-to-bed ratios, more advanced nurse skill mix, greater hospital volume, and others have been shown to reduce failure to rescue rates. In addition, use of technology-supported interventions (such as patient monitoring systems and rapid response teams), standardized communication tools, or simulation training can improve timely recognition and response to clinical deterioration and reduce failure to rescue.
- The developer states that this measure is a respecified version of CBE#0353 - Failure to Rescue 30-day Mortality, which is no longer endorsed. It also is intended to replace the CMS PSI 04, which is currently being used in the Hospital Inpatient Quality Reporting (HIQR) Program. After the measure was submitted to Battelle, the developer added more information in response to the staff assessment, indicating that CMS opted to not finalize retirement of PSI 04 and committed to redesigning PSI 04 to address the concerns of hospitals and health care providers, while retaining the key quality concept underlying the measure (e.g., failure to rescue, or death, of a patient who experienced a significant postoperative complication).
- The developer reports initial risk-standardized rates for the measure across 2,055 facilities with 25 qualifying records. The mean is 46.2 with an interquartile range of 29.33 to 60.95.
- The developer did not provide direct patient input for this measure but does note the communications received from the patient community with respect to the retirement of the PSI 04. The developer states that these communications clearly articulated the perceived value of CMS PSI 04 as a broad measure of postoperative mortality and hospitals’ skill at rescuing patients who experience complications.
Feasibility Acceptance
Strengths:
- The developer did not conduct a feasibility assessment, stating that “Because this measure is based on readily available administrative claims data, feasibility is not an issue." Developers add that PSI 04, a measure with similar design, has been used by CMS for more than a decade and that "No difficulties have been reported with respect to data collection, availability of data, missing data, timing and frequency of data collection, sampling, patient confidentiality, or time and cost of data collection. Hospitals routinely generate and transmit claims in a timely manner for all Medicare beneficiaries."
- There are no fees associated with use of this claims-based measure. The measure specifications will be available upon request through the CMS QualityNet Help Desk.
Limitations:
- None
Rationale:
- The developer did not conduct a feasibility assessment, stating that “Because this measure is based on readily available administrative claims data, feasibility is not an issue." Developers add that PSI 04, a measure with similar design, has been used by CMS for more than a decade and that "No difficulties have been reported with respect to data collection, availability of data, missing data, timing and frequency of data collection, sampling, patient confidentiality, or time and cost of data collection. Hospitals routinely generate and transmit claims in a timely manner for all Medicare beneficiaries."
- There are no fees associated with use of this claims-based measure. The measure specifications will be available upon request through the CMS QualityNet Help Desk.
Scientific Acceptability
Strengths:
- Measure is well-defined and specified.
- Accountable entity-level reliability was assessed with signal-to-noise analysis performed on 2019-2020 data with 1,087,624 patients across 2,055 entities. A decile table of reliability by population size was provided with a median reliability of 0.568. Approximately 45-50% of entities have a reliability >0.6.
Limitations:
- Approximately 50-55% of entities have reliability less than the threshold of 0.6.
Rationale:
Majority of entities have a reliability <0.6. Consider mitigation for entities with low denominator size. some possible mitigation strategies to improve these estimates could be to:
- Empirical approaches outlined in the report, MAP 2019 Recommendations from the Rural Health Technical Expert Panel Final Report, https://www.qualityforum.org/WorkArea/linkit.aspx?LinkIdentifier=id&ItemID=89673.
- Consider a higher minimum case volume.
- Extend the time frame.
- Focus on applying mitigation at the lower volume providers.
Strengths:
- The developer conducted face and empiric validity testing of the measure score (i.e., accountable entity-level).
- For face validity, the developer notes that it obtained input from members of the TEP to determine whether they think the measure as specified will help inform consumers and help providers improve quality. However, the expertise of the TEP members was not disclosed. Nine of the 10 members (90%) voted “yes” that implementation of the measure in hospital inpatient quality reporting programs (in place of current PSI 04) is likely to lead to improve quality of care by reducing the frequency of failure to rescue. The one member who disagreed felt that the proposed denominator expansion (adding patients who experience less serious complications after surgery) makes the measure less relevant to identifying hospitals’ performance in rescuing higher risk/serious cases.
- For empiric testing, the developer conducted convergent validity testing by comparing hospital-level failure-to-rescue rates with rates of risk-standardized 30-day readmission and mortality rates (e.g., hospital-wide unplanned all-cause readmissions), complications for hip/knee replacement patients and a composite measure of patient safety and adverse events. The developer did not expect strong correlations because patient safety is a complex construct, and these measures differ from the failure-to-rescue measure in terms of the populations and conditions being measured. Correlations were weak, but are stronger for proposed measure compared to the PSI 04 measure.
- The developer also conducted construct validity testing, hypothesizing that failure-to-rescue rates would be lower at major teaching hospitals, urban hospitals, and hospitals with high nurse staffing and skill mix than at non-teaching hospitals, rural hospitals, and hospitals with low nurse staffing and skill mix, respectively. The results support these hypotheses for all “known groups” except rural/urban location. However, the developer does not provide a rationale as to why.
- Risk adjustment: The measure is risk-adjusted for 126 factors. The developer explored social risk factors, but it did not include them in the final model. The developer states this is due to the empirical analyses confirming the measure is neutral across social risk groups, including race, ethnicity, urbanicity/rurality, and sex, due to adjustment for all other 126 patient characteristics. The c-statistic for the final logistic model is 0.818.
- After the measure was submitted to Battelle, the developer added more information in response to the staff assessment: The TEP was composed of clinicians from a range of specialties, health care quality subject matter experts, and three patient/caregiver representatives.
Limitations:
- None
Rationale:
- The developer conducted face and empiric validity testing of the measure score (i.e., accountable entity-level).
- For face validity, the developer notes that it obtained input from members of the TEP to determine whether they think the measure as specified will help inform consumers and help providers improve quality. However, the expertise of the TEP members was not disclosed. Nine of the 10 members (90%) voted “yes” that implementation of the measure in hospital inpatient quality reporting programs (in place of current PSI 04) is likely to lead to improve quality of care by reducing the frequency of failure to rescue. The one member who disagreed felt that the proposed denominator expansion (adding patients who experience less serious complications after surgery) makes the measure less relevant to identifying hospitals’ performance in rescuing higher risk/serious cases.
- For empiric testing, the developer conducted convergent validity testing by comparing hospital-level failure-to-rescue rates with rates of risk-standardized 30-day readmission and mortality rates (e.g., hospital-wide unplanned all-cause readmissions), complications for hip/knee replacement patients and a composite measure of patient safety and adverse events. The developer did not expect strong correlations because patient safety is a complex construct, and these measures differ from the failure-to-rescue measure in terms of the populations and conditions being measured. Correlations were weak, but are stronger for proposed measure compared to the PSI 04 measure.
- The developer also conducted construct validity testing, hypothesizing that failure-to-rescue rates would be lower at major teaching hospitals, urban hospitals, and hospitals with high nurse staffing and skill mix than at non-teaching hospitals, rural hospitals, and hospitals with low nurse staffing and skill mix, respectively. The results support these hypotheses for all “known groups” except rural/urban location. However, the developer does not provide a rationale as to why.
- The measure is risk-adjusted for 126 factors. The developer explored social risk factors, but it did not include them in the final model. The developer states this is due to the empirical analyses confirming the measure is neutral across social risk groups, including race, ethnicity, urbanicity/rurality, and sex, due to adjustment for all other 126 patient characteristics. The c-statistic for the final logistic model is 0.818.
Equity
Strengths:
- Developer evaluated disparities by race, ethnicity, sex, age, and findings reported are adjusted by all risk adjustment model factors, which includes age, comorbidities, principal diagnosis, and in-hospital morbidity risk (Elixhauser index)
- No differences by race, ethnicity, age, or sex were found in risk-adjusted analyses.
Limitations:
- None
Rationale:
- Developer evaluated disparities by race, ethnicity, sex, and age using risk-adjusted models; no disparities were found. Risk adjustment included age, comorbidities, principal diagnosis, and Elixhauser index.
Use and Usability
Strengths:
- Developer indicates that the measure is planned for use in public reporting.
- Developer cited evidence from several studies, including two systematic reviews, to highlight hospital-level strategies entities can implement to improve performance meant to enhance the timely identification of clinical deterioration and treatment. Promising strategies included improved nurse training and staffing levels/patterns (14 studies), use of early warning systems and checklists (8 studies), nursing monitoring and surveillance (6 studies), improved documentation, use of escalation protocols, patient monitoring systems, and rapid response teams.
- After the measure was submitted to Battelle, the developer added more information in response to the staff assessment: The measure has been designed and tested to replace CMS PSI 04, which is currently being used in the Hospital Inpatient Quality Reporting (HIQR) Program (00134-02-C-HIQR, formerly CBE #0351).
Limitations:
- None
Rationale:
- Developer indicates that the measure is planned for use in public reporting but does not provide any other information such as program name, purpose, geographic coverage, level of analysis, etc. Developer suggests several strategies entities could implement to enhance timely identification of clinical deterioration and treatment, and improve performance on the measure, are improved nurse staffing, simulation training, communication tools, monitoring/warning systems, and rapid response systems; however, no details regarding how and where similar tools have been implemented is provided.
Summary
N/A
Committee Independent Review
Measure Summary
Importance
Agree with staff assessment.
Feasibility Acceptance
Agree with staff assessment. I appreciate measures collected via claims to reduce reporting burden.
Scientific Acceptability
Agree with staff assessment.
Agree with staff assessment.
Equity
Agree with staff assessment.
Use and Usability
Agree with staff assessment. I appreciate the examples of improvement opportunities to identify clinical deterioration.
Summary
This measure supports the significance of identifying and treating clinical deterioration following a surgical procedure. Monitoring this and implementing improvement opportunities will improve outcomes and prevent potential harm for patients.
Important for monitoring clinical deterioration
Importance
Agree with staff assessment. Strong evidence base and rationale to update an existing measure to address multiple stakeholder concerns. Data provided showing this meausre can result in actionable change.
Feasibility Acceptance
Agree with staff assessment - Measure based on readily available administrative claims data, no fees associated with measure. No difficulties reported in prior verson of similiar measure.
Scientific Acceptability
Agree with staff assessment. A decile table of reliability by population size showed a median reliability of 0.568. Approximately 45-50% of entities have a reliability >0.6.
Agree with staff assessment. Good face validity based on feedback from TEP members. Empiric validity results supported.
Equity
Agree with staff assessment - Conducted a social disparities analysis and found no significant differences by race, ethnicity, age or sex.
Use and Usability
Agree with staff assessment - for public use. Would benefit from implementation plan.
Summary
Support for using this measure for timely identification of clinical deterioration and devising strategies for treatment of preventable complications,
Failure to rescue monitoring
Importance
Agree with staff assessment. Believe this measure can be used for quality improvement efforts in reducing surgical death within 30 days. This measure could be a change agent.
Feasibility Acceptance
Agree with staff assessment. Data is readily available, no fees and similar measure used for more than a decade without problems.
Scientific Acceptability
Agree with staff assessment. I worry about rural hospitals. I believe 50%-55% reliability less than the threshold is not acceptable .
Agree with staff assessment.
Equity
Agree with staff assessment. Evaluate it for SDOH and race, ethnicity, age, and sex.
Use and Usability
Agree with staff assessment. It will be publicly reported. Transparency of information is important.
Summary
NA
n/a
Importance
Agree with staff, would add that some of this underlying literature about association between staffing etc and anesthesiologists is very old (30 years) before the development of code teams. The feasibility analysis however is more recent giving face validity.
Feasibility Acceptance
Agree with staff, no need for feasibility assessment in this context
Scientific Acceptability
Agree with staff
Agree with staff including proposal to address with mitigation strategies
Equity
Agree with staff, no differences found despite checking
Use and Usability
Agree with staff
Summary
n/a
Failure to Rescue
Importance
Agree with assessment, meets need of a retired measure. Addresses identified problem of in-hospital mortality and provides actionable items which may effect positive change.
Feasibility Acceptance
Agree with assessment, previous measure PSI 04 has shown this is a feasible metric with readily obtainable data and ability to report findings in a timely manner.
Scientific Acceptability
There are valid concerns pertaining to institutions which may not achieve the threshold for measure reliability. Interventions recommended by the staff assessment to extend time frame or increase minimum case threshold may not achieve desired outcomes and may further complicate interpretation of the data.
There are many confounding factors within this measure based on the variability of the patient being assessed. Correlation was not expected to be strong, but did outperform PSI 04. Interested in details about TEP selection and opinion which led to their vote of validity (90%) vs the one who dissented.
Equity
When adjusted for additional risk factors and comorbidities no disparities were identified.
Use and Usability
Agree with staff, the measure is proposed to replace PSI 04. It will be reported publicly. Clear interventions exist to improve institutional performance such as mandated nurse to patient ratios, implementation of best practices including rapid response teams and specialty training, and the utilization of technology to better monitor patients in real time.
Summary
As PSI 04 is retired this is a timely metric to continue evaluating institutional performance in mitigating serious adverse events and complications which may occur. Clear interventions exist and have been identified by the measure authors which adequate and strong evidence in support.
Summary
Importance
Agree with staff comments.
Feasibility Acceptance
Agree with staff comments.
Scientific Acceptability
Reliability of less than 0.6 for 50-55% of facilities needs to be addressed.
Agree with staff comments.
Equity
Agree with staff comments.
Use and Usability
Agree with staff comments.
Summary
n/a
Failure to rescue - surgical patients
Importance
Agree with staff assessment. Glad to see CMS aiming to bring forth a measure that may prove to be more representative as well as a measure that hospital teams are better able to impact.
Feasibility Acceptance
Agree that the feasibility assessment may not be needed since the measure is modeled after PSI 04
Scientific Acceptability
Agree with staff assessment.
Agree with staff assessment, and measure developer appears to have met all the listed criterion in the guidebook.
Equity
Measure can be stratified to identify issues that potentially have equity implications.
Use and Usability
Developer provided the potential settings where the measure can be utilized. Similarly to PSI 04, this new measure likely does not require additional usability ratings. Eventually important to define how/what settings the measure can be used as it has the potential to impact care beyond just acute care hospitals.
Summary
Important measure to continue to test and refine as needed. The current PSI 04 is controversial for large referral centers and perhaps living up to its intent.
Summary
Importance
Agree with staff assessment.
Feasibility Acceptance
Agree with staff assessment.
Scientific Acceptability
Agree with staff assessment but I am uncertain if their numbers will improve.
I believe the scientific acceptability testing mets expectation.
Equity
Agree with staff assessment.
Use and Usability
Agree with staff assessment.
Summary
It is an important measure, and it was designed to replace CMS PSI 04, which is currently being used in the HIQR Program.
N/A
Importance
I agree with the Staff's assessment.
Feasibility Acceptance
I agree with the Staff's assessment.
Scientific Acceptability
I agree with the Staff's assessment.
I agree with the Staff's assessment.
Equity
I agree with the Staff's assessment.
Use and Usability
I agree with the Staff's assessment.
Summary
N/A
Agree with merits of the…
Importance
Agree with staff.
Feasibility Acceptance
Agree with staff.
Scientific Acceptability
Concern around patients who are not DNR but decline selected services which could otherwise contribute to a successful numerator/outcome, particularly in the higher end of this age range. Could consider reducing the upper age limit of eligibility.
Agree with staff.
Equity
Appropriate inclusion into measure logic.
Use and Usability
Agree with staff.
Summary
Agree with merits of the measure and benefits of employing a system for identifying and managing surgical complications early and aggressively.
Summary
Importance
The seminal research papers (1992, 1997) raised awareness that modifiable factors (e.g., nurse/bed ratios, MRI facilities, bone marrow transplant units, residency training programs, higher nurse staffing, better nursing skill) were associated with lower failure to rescue (FTR) rates. These findings were further validated by multiple studies (2007-2015). A 2015 systematic review also identified several hospital characteristics associated with delayed escalation of care and higher FTR rates.
Feasibility Acceptance
This was a bit of a judgement call. No feasibility testing was done. The rationale was, “Because this measure is based on readily available administrative claims data, feasibility is not an issue. A similarly designed measure (CMS PSI 04) has been used by CMS for over a decade. No difficulties have been reported with respect to data collection, availability of data, missing data, timing and frequency of data collection, sampling, patient confidentiality, or time and cost of data collection. Hospitals routinely generate and transmit claims in a timely manner for all Medicare beneficiaries.”
It will be interesting if this will be precedent when other “new” measures are really improved measures that are expected to be replacement of prior widely used measure by CMS.
Scientific Acceptability
Specifications were well-defined and easily replicable. All data came from the fields available on Medicare FFS claims and Medicare Advantage shadow claims (inpatient encounter records), including ICD-10-CM diagnosis codes for comorbidities present on admission, ICD-10-CM principal diagnosis codes, ICD-10-PCS procedure codes.
At the accountable entity-level, signal-to-noise reliability was estimated as an intraclass correlation coefficient based on a two-way mixed model with facility random effects (C,1). The data sources were 2019-2020 Medicare claims and deaths. The sample was with 1,087,624 patients across 2,055 entities.
Reliability was moderate. The median was 0.568. Thus, about 45-50% of entities were above 0.6, and about 50-55% were below this threshold.
I appreciated the staff’s recommendation on how to address limitations, citing Empirical approaches outlined in the report, MAP 2019 Recommendations from the Rural Health Technical Expert Panel Final Report. These made sense to me.
Validity was examined using three approaches.
Convergent validity refers to the degree to which multiple measures of a single underlying concept are positively correlated with each other. To assess the convergent validity of the measure, measure results were compared results from related measures of patient safety and outcomes.
For all but one of these comparisons, the proposed measure demonstrates higher convergent validity than the current CMS PSI 04 measure.
Examining variation by “known groups”. Their findings are also consistent with findings from prior research described earlier. “The data support these hypotheses for all “known groups” except rural/urban location.”
Face validity results are as follows:
- 9 of 10 members (90%) voted “yes” that the measured outcome (rate of 30-day mortality among surgical inpatients with complications) provides a representation of relevant quality in a facility.
- 9 of 10 members (90%) voted “yes” that implementation of the measure in hospital inpatient quality reporting programs (in place of current PSI 04) is likely to lead to improve quality of care by reducing the frequency of failure to rescue.
- 5 of 5 members (100%) who are employed by a “measured entity” (i.e., employed or affiliated with hospital organizations) voted “yes” that the proposed measure is easy to understand and may be useful for decision-making.
Equity
Method for risk adjustment was appropriate. The final risk-adjustment model was estimated using cluster-adjusted multivariable logistic regression to optimize calibration, after testing both logistic and probit link functions. The covariates were age, discharge quarter, Modified Diagnosis-Related Groups (MDRGs), AHRQ’s default Clinical Classifications Software Refined (CCSR) for International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM)-codes, applied to the principal diagnosis on each record, and Elixhauser Index for Risk of In-hospital Mortality. The measure is risk-adjusted for 126 factors. The developer explored social risk factors, but it did not include them in the final model.
No differences by race, ethnicity, age, or sex were found in risk-adjusted analyses
Use and Usability
To be used for public reporting. The measure has been designed and tested to replace CMS PSI 04, which is currently being used in the Hospital Inpatient Quality Reporting (HIQR) Program (00134-02-C-HIQR, formerly CBE #0351).
Summary
This measure was designed and tested to replace CMS PSI 04, which is currently being used in the Hospital Inpatient Quality Reporting (HIQR) Program (00134-02-C-HIQR, formerly CBE #0351). Decision to improve the original measure by CMS was in response to public feedback. The measure was further revised by research team at the Center for Healthcare Policy and Research at University of California, Davis, work that was contracted by CMS. The current measure is an updated and completely re-tested version of a previously CBE-endorsed measure #0353, “Failure to Rescue 30-day Mortality.” This measure was stewarded by Silber and colleagues at the Children’s Hospital of Philadelphia (CHOP) and used extensively by the research and quality improvement communities. CHOP allowed CBE endorsement to lapse in 2021. It was interesting that feasibility testing was not done given established feasibility on original measure. Scientific acceptability seemed to fall within the usual standards for QM testing. Risk adjustment was extensive and variation by socio’s following adjustment were not significant. The inclusion criteria include enrollment in Medicare, but the denominator includes patients aged 18 years and older—I found this a bit odd since most Medicare beneficiaries are 65 years or older with two exceptions (end stage renal, ALS). Nevertheless, this team further improved a measure that has been widely in use and more than three decades of research to support its significance.
N/A
Importance
Agree with staff preliminary assessment. I participated as a TEP member for this measure and I disclosed in my COI.
Feasibility Acceptance
Agree with staff preliminary assessment. I participated as a TEP member for this measure and I disclosed in my COI.
Scientific Acceptability
Agree with staff preliminary assessment. I participated as a TEP member for this measure and I disclosed in my COI.
Agree with staff preliminary assessment. I participated as a TEP member for this measure and I disclosed in my COI.
Equity
Agree with staff preliminary assessment. I participated as a TEP member for this measure and I disclosed in my COI.
Use and Usability
Agree with staff preliminary assessment. I participated as a TEP member for this measure and I disclosed in my COI.
Summary
N/A
Measure meets all criteria
Importance
There are a number of facility level interventions that can be instituted to improve rates.
Feasibility Acceptance
All data are from administrative claims.
Scientific Acceptability
Though the reliability results were strong, the developers note that you can expect higher FTR rates with lower hospital volume, lower nurse staffing, and non-teaching status. This seems to disadvantage rural hospitals, but seems to have been accounted for in the risk adjustment model.
No comments
Equity
The developers reported the rates across different racial and ethnic groups and explained why these variables were not included in the risk adjustment model.
Use and Usability
Measure will be used in a public reporting program.
Summary
See above.
The evaluation of failure-to…
Importance
Measuring and reporting failure to rescue is an important endeavor.
Feasibility Acceptance
The measure appears feasible given the data collected.
Scientific Acceptability
Developers report low reliability across multiple hospitals. The use of claims data to measure failure to rescue is inherently problematic given our inability to capture all the underlying clinical factors affecting a patient's likelihood of experiencing a poor outcome.
Agree with staff assessment.
Equity
Agree with staff assessment.
Use and Usability
Developer states this will be a publicly reported measure however provides no further details. I have considerable concerns regarding the consequences of this measure being publicly reported when it has the data flaws noted above.
Summary
The evaluation of failure-to-rescue is important. However, this measure has several concerning flaws. First, the developers describe low reliability in their evaluation of the measure. Second, not adjusting for social factors seems problematic as they likely impact failure-to-rescue. Third, measuring death within 30 days regardless of location seems far too broad. Outcomes of interest should be procedure specific, such as developing an MI after a major abdominal operation, and not getting hit by a car when crossing the street 3 weeks after an abdominal operation.
Final Comment
Importance
Agree with the staff assessment.
Feasibility Acceptance
Agree with the staff assessment.
Scientific Acceptability
Agree with the staff assessment.
Agree with the staff assessment.
Equity
Agree with the staff assessment.
Use and Usability
Agree with the staff assessment. But add this measure should be reworked to be used for all payers not just Medicare. If it planned for use in the public domain, too much risk adjustment. Great addition to an internal QI program but concerned if a facility does not have a sizable Medicare population, the measure might not be representative of the entire population.
Summary
Why only Medicare? If time and money are going to be used to develop measures, they should be for all payers.
N/A
Importance
I agree with staff assessment though I would like to know why Silber and CHOP have let the similar CBE lapse in 2021.
Feasibility Acceptance
I agree with staff assessment.
Scientific Acceptability
I agree with staff assessment.
I agree with staff assessment.
Equity
I agree with staff assessment.
Use and Usability
I agree with staff assessment.
Summary
Reasonable measure though understanding why a previous similar measure was lapsed would be helpful
Important Measure
Importance
Failure to recognize/rescue is a hot topic and critically important. The logic models proposed make sense and the number of factors that can influence this patient outcome are well documented and described.
Feasibility Acceptance
Agree with staff assessment
Scientific Acceptability
Agree with staff assessment
Agree with staff assessment
Equity
The authors have done a thorough review of potential equity concerns and found none.
Use and Usability
Agree with staff assessment
Summary
Almost ready for prime time
Support this measure moving…
Importance
Agree with staff assessment.
Feasibility Acceptance
Agree with staff assessment.
Scientific Acceptability
Agree with staff assessment.
Agree with staff assessment.
Equity
Agree with staff assessment.
Use and Usability
Agree with staff assessment.
Summary
Support this measure moving forward to enhance capture of critical health outcome.
Public Comments
Public Comments from Pre-Rulemaking Measure Review (PRMR)
CBE #4125 - Thirty-day Risk Standardized Death Rate among Surgical inpatients with Complications (Failure-to-Rescue) is also a measure under consideration for potential inclusion in the Hospital Inpatient Quality Reporting Program (HIQR) as MUC2023-049 and is currently undergoing review by the Pre-Rulemaking Measure Review (PRMR) committees. Prior to its review, the measure was posted for PRMR public comment, and received 11 comments, which can be found here: https://p4qm.org/sites/default/files/2024-01/Compiled-MUC-List-Public-Comment-Posting.xlsx. Please review and consider these PRMR comments for MUC2023-049 in addition to any submitted within the public comment section of this measure’s webpage. If there are no comments listed in the public comment section of this webpage, then none were submitted.