The American Journal of Surgery (2014) 207, 326-330
Midwest Surgical Association
Medicare post-discharge deaths and readmissions following elective surgery Donald E. Fry, M.D.a,b,c,*, Michael Pine, M.D, M.B.A.a,d, Gregory Pine, B.A.a a
Michael Pine & Associates, 1 East Wacker Drive, #1201, Chicago, IL 60601, USA; bDepartment of Surgery, Northwestern University Feinberg School of Medicine, Chicago, IL, USA; cDepartment of Surgery, University of New Mexico School of Medicine, Albuquerque, NM, USA; dDepartment of Medicine, The University of Chicago, Chicago, IL, USA
KEYWORDS: Risk-adjusted outcomes; Elective surgical care; Hospital readmissions; Postdischarge death rates; Control charts; Postoperative mortality rates
Abstract BACKGROUND: The frequency of 90-day, postdischarge deaths and readmissions in Medicare patients undergoing elective surgical procedures has not been well studied. METHODS: The Medicare MedPar database for 2009 to 2010 was used to develop inpatient risk-adjusted, postoperative length-of-stay (RApoLOS) prediction models for live discharges in 21 categories of elective operations. Moving average control charts were used in each category to define RApoLOS outliers (.3s). The relationships between RApoLOS outliers and all postdischarge deaths and readmissions within 90 days of discharge were assessed. RESULTS: The inpatient mortality rate was .5%. Of 2,054,189 live discharges, 147,292 (7%) were RApoLOS outliers. There were 14,657 postdischarge deaths (.7%) and 187,566 readmissions (9%). RApoLOS outliers had a 3.5% death rate and a 17% rate of readmission, while those found not to be RApoLOS outliers had a .5% death rate and a 9% readmission rate (P , .0001). CONCLUSIONS: RApoLOS outliers have increased rates of postdischarge deaths and readmissions. Ó 2014 Elsevier Inc. All rights reserved.
Complications of inpatient care of medical and surgical patients are identified as a cause of deaths, patient morbidity, and excess costs to the health care system. Many reports identify deaths and complications that occur only during the actual inpatient phase of patient management, and postdischarge events are not reported in large part because of the difficulty in getting accurate postdischarge data. Recent data indicate that as many as one third of hospitalized Medicare patients are readmitted to the hospital ,90 days after an
The authors declare no conflicts of interest. * Corresponding author. Tel.: 11-312-467-1700; fax: 11-312-4671705. E-mail address: [email protected] Manuscript received July 14, 2013; revised manuscript September 1, 2013 0002-9610/$ - see front matter Ó 2014 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.amjsurg.2013.09.007
inpatient episode of care.1 Many of these readmissions follow emergency medical and surgical hospitalizations. A plausible argument can be made that postdischarge readmission of Medicare patients may be driven by the progression of underlying disease risk factors and that these readmissions may not be avoidable. Others contend that readmissions are a direct consequence of the index hospitalization and should be viewed as complications of care. All agree that readmissions are expensive to the health care system. With major inpatient elective operations, postdischarge complications are a source of concern, but the actual frequency of these events has been difficult to define. Postdischarge events are not systematically reported, and even after elective procedures, patients may be readmitted to hospitals other than the sites of their index operations. The costs of readmission have reached a significant level of
D.E. Fry et al.
Postdischarge deaths and readmissions
concern, resulting in the implementation of financial penalties by the Centers for Medicare and Medicaid Services for those rates of readmissions that are deemed to be excessive.2 In this study, we evaluated postdischarge deaths and readmissions to acute care hospitals among a large population of Medicare patients undergoing a broad array of different elective surgical procedures. We tested the hypothesis that major adverse outcomes of the index hospitalization are a predictor of 90-day postdischarge death and readmission.
Methods We used the MedPar database from Medicare for 2009 and 2010 to evaluate the postdischarge events of patients undergoing 21 different categories of elective surgical procedures. The categories were 2 neurosurgery; 1 face, head, and neck; 1 lung and thorax; 2 cardiac, 3 vascular; 2 gastrointestinal; 1 miscellaneous high risk; 1 laparoscopic cholecystectomy; 1 hepatobiliary and pancreatic; 1 abdominal wall hernia; 1 renal and hysterectomy; 1 prostate and urinary; 1 female reproductive (not hysterectomy); 1 breast; 1 vertebral surgery; and 1 hip and knee replacement. Multiple categories were defined in neurosurgery, cardiac surgery, vascular surgery, and gastrointestinal surgery to optimize uniformity in anticipated lengths of stay and death rates. For example, carotid endarterectomy has a very different length of stay and mortality rate compared with aortic aneurysm repair within vascular surgery. This Medicare data set permits the identification of administrative claims for all hospitalizations and permits the identification of all readmissions. It does not have Part B claims for physicians and ambulatory services. Dates of death for Medicare beneficiaries are identified and permit the determination of 90-day postdischarge mortality rates.
Defining the analytic database All cases within each surgical category were identified using International Classification of Diseases, Ninth Revision, codes. Only elective patients aged R65 years who underwent operations on days 0, 1, or 2 of hospitalization were included from hospitals with R20 cases for the study period. Inpatient deaths were identified and removed from the subsequent postoperative length-of-stay (poLOS) analysis. Present-on-admission quality screens were used to eliminate poor coding hospitals for model development, but these facilities were retained for all outcome analyses.3
Risk-adjusted poLOS outliers Previously we have identified risk-adjusted (RApoLOS) outliers to be patients with clinically relevant major complications after inpatient surgical operations.4,5 RApoLOS outliers refine the definition of a relevant complication of care in that they have .3s observed length of stay compared with predicted, and they have costs that are
327 statistically greater than patients with coded complications who are not RApoLOS outliers. To define these patients, a standard list of risk factors for clinical diagnoses present on admission for each of the 21 surgical groups was defined. A preliminary forward stepwise linear regression model for poLOS was then computed from those patients within each category who had no coded complications. Coefficients were determined for significant risk factors (P , .05). A unique prediction model was derived for each category of elective procedure. Hospital variables were used to eliminate these effects on coefficients of risk models. A moving average control chart by methods previously published was then created within each category of operation and within each hospital.6,7 The observed poLOS was identified, and the predicted poLOS was calculated from the specific prediction model. Within each hospital for each eligible procedure, the total predicted poLOS days were set equal to the total observed days by multiplication of predicted values by a constant. The observed-minuspredicted poLOS differences for all patients within each category were then determined in the temporal sequence within each hospital. Within each hospital, observedminus-predicted differences for each patient in the temporal sequence were subtracted from each other to give an absolute difference. The mean of all absolute differences within each hospital for each category was computed, and this mean was multiplied by a constant of 2.66 to define the 3s upper control limit of the control chart.8 All categories of procedures had outliers with dramatically extended poLOS. These dramatic outliers had a profound impact on the mean of absolute differences between sequential cases within a hospital. After the first control chart was completed, the outlier cases were removed, and the process was repeated with the revised mean of absolute differences to remove a second group, albeit smaller, number of outliers. The process was iteratively repeated until no outliers remained. In statistical process control terms, the remaining cases had only common-cause variation, and all remaining cases reflected an ‘‘in control’’ stable system. Total RApoLOS is the number of cases that needed to be removed to establish the stable system. All prediction models were then further refined by recalibration of the coefficients of the prediction model for each of the 21 categories, but this time only with patients who were not outliers. This enhanced the accuracy of overall models by removing the impact of outlier cases on final coefficients. This final RApoLOS model was then run against all nonoutlier cases from the preliminary evaluation to identify additional outliers after recalibration of coefficients. Rates of RApoLOS outliers were then computed for each of the 21 categories of operations.
Postdischarge deaths and readmissions All live discharges from the index hospitalization were then examined for 90 days after discharge. All 90-day
328
The American Journal of Surgery, Vol 207, No 3, March 2014
Table 1 Observed deaths and readmissions of all patients divided into those who were RApoLOS outliers and who had coded complications in the discharge abstract No RApoLOS outliers Variable No. of patients: (% of all patients) Postdischarge deaths (death rate) Odds ratio No. of readmissions (% of total patients) Odds ratio
RApoLOS outliers
No coded complications
Coded complications
No coded complications
Coded complications
Total patients
1,141,937 (55.6%)
764,960 (37.2%)
24,067 (1.2%)
123,225 (6.0%)
2,054,189
3,800 (.33%)
5,776 (.76%)
542 (2.28%)
4,557 (3.70%)
14,675 (.71%)
1.0 85,397 (7.48%)
2.3 77,592 (10.1%)
6.9 3,913 (16.3%)
11.5 20,664 (16.8%)
187,566 (9.13%)
1.0
1.4
2.4
2.5
RApoLOS 5 risk-adjusted, postoperative length of stay.
readmissions to acute care hospitals were identified for each category. It is appreciated that all readmissions may not be directly linked to the index hospitalization, especially in a Medicare population that may have a number of unrelated acute and chronic illnesses that may necessitate rehospitalization. For this initial analysis, we chose to include all readmissions because our preliminary observations led us to believe that 90% of readmissions within 90 days are linked to the index hospitalization. Death records were then queried to identify all live discharges who died during the 90 days after discharge. All computations and data analysis were performed using SAS version 9.1.3 (SAS Institute Inc, Cary, NC).
Results After all inpatient deaths from the index hospitalization were removed (.5%), a total of 2,054,189 live discharges were evaluated for postdischarge events. Within 90 days of discharge from the hospital, there were 14,675 deaths (.7%) and 187,566 readmissions (9%). Among live discharges, a total of 888,185 (43%) had R1 coded complications for the index hospitalization. A total of 147,292 patients (7%) were RApoLOS outliers. Table 1 lists the data relationships among RApoLOS outliers, identified coded complications, and 90-day postdischarge and readmissions. It should be emphasized that 37% of all discharges had coded complications but were not RApoLOS outliers. Also of note, 24,067 of all RApoLOS outliers (16%) did not have coded complications at discharge. Coded complications and RApoLOS were independently significant in association with postdischarge deaths and readmissions (P , .0001). However, patients with no coded complications who were RApoLOS outliers had a significantly higher rate of postdischarge death and readmission than did those patients with coded complications who were not RApoLOS outliers. Odds ratios predicting postdischarge death or readmission were highest among the patients who had coded complications and were RApoLOS outliers by univariate analysis (P , .0001).
RApoLOS outliers had a 3.5% postdischarge death rate and a 17% rate of readmission, while nonoutliers had a .5% death rate and a 9% readmission rate (P , .0001). Table 2 identifies rates of coded complications and RApoLOS outliers of specific high-volume operations. The 90-day postdischarge deaths and readmission rates for cardiac and colon surgery were higher than the mean for the entire patient population, while total joint replacement rates were lower.
Comments In this study, we examined elective surgical procedures that spanned a broad scope of severity and costs. The results indicate that more patients died in the 90 days after discharge than died as inpatients after the operations and that overall, 9% of these Medicare patients were readmitted to the hospital in this 90-day period. Not surprisingly, readmission rates were much higher for major cardiothoracic and abdominal operations. We have identified that RApoLOS outliers were statistically associated with 90-day postdischarge deaths and readmissions (P , .0001). These observations are consistent with our several publications across different elective surgical procedures that RApoLOS outliers represent significant inpatient morbidity and, on the basis of the current data, are associated with significant postdischarge adverse events. RApoLOS outliers who do not have any coded complications are of considerable interest. These outliers predict postdischarge deaths and readmissions with nearly the same statistical significance as the outliers with coded complications. The excess length of stay is unlikely to be the consequence of addition unnecessary days of hospitalization, because the Inpatient Prospective Payment System of the Centers for Medicare and Medicaid Services provides a negative incentive for extended hospitalization. It is likely that major complications of care have been omitted from the abstract summaries of these patients. In elective surgical care, the excess poLOS should seldom be driven by poor discharge planning or disposition issues. Administrative
D.E. Fry et al. Table 2 surgery
Postdischarge deaths and readmissions
329
Data on RApoLOS outliers, 90-day readmissions, and 90-day deaths that occurred in cardiac, joint replacement, and colon
Variable
Cardiac
Total joint replacement
Colon surgery
All patients
Inpatient mortality rate Total live discharges Index hospitalization No CC, no RApoLOS CC, no RApoLOS No CC, RApoLOS CC, RApoLOS Postdischarge deaths No CC, no RApoLOS CC, no RApoLOS RApoLOS Total readmissions No CC CC, no RApoLOS No CC, RApoLOS CC, RApoLOS
2.42% 155,596
.11% 682,459
2.72% 69,986
.54% 2,054,189
29,566 (19.0%) 108,052 (69.4%) 395 (.25%) 17,583 (11.3%) 1,929 (1.24%) 134 890 905 24,055 (15.5%) 3,672 16,996 69 3,318
327,434 (48.0%) 323,440 (47.4%) 3,671 (.54%) 27,914 (4.09%) 1,022 (.15%) 309 484 229 36,893 (5.41%) 14,868 18,553 289 3,183
23,065 (33.0%) 39,850 (56.9%) 238 (.34%) 6,833 (9.76%) 2,115 (3.02%) 277 1,326 512 10,018 (14.3%) 2,583 6,397 27 1,011
1,141,937 (55.6%) 764,960 (37.2%) 24,067 (1.2%) 123,225 (6.0%) 14,675 (.71%) 3,800 5,776 5,099 187,566 (9.13%) 85,397 77,592 3,913 20,664
CC 5 coded complications; RApoLOS 5 risk-adjusted, postoperative length of stay. Twenty-one categories of Medicare patients undergoing elective surgical procedures were evaluated. Risk-adjusted length-of-stay outliers from the index hospitalization predicted increased rates of 90-day postdischarge deaths and readmissions to the hospital.
issues in patient disposition would not explain the higher rates of deaths and readmissions after discharge. Another interesting population of patients is those with coded complications who are not RApoLOS outliers. These patients have higher readmission rates than those who do not have any coded complications. In the Medicare population, many of these coded complications may be driven by incentives for higher Medicare severity-diagnosisrelated group payments. The RApoLOS methodology does provide a measure of severity to the coded complications that can be of value in the identification of those patients in need of focused review of processes of care and those at greatest risk for readmission. There were 3 limitations of this study. First, all-cause deaths and readmissions are reported, and some may not be related to the index hospitalizations or operations. Interval trauma or unrelated diseases may account for a portion of the 90-day postdischarge events. Second, the absence of Part B data in this data set prevents a total cost analysis of the index and readmission hospitalizations. Finally, risk adjustment using administrative data is always challenged, and models with greater clinical data are always desirable. Accurate present-onadmission coding and the incorporation of readily available clinical elements such as admission laboratory data into the administrative data set have the prospect of enhancement of modeling without resorting to the more expensive process of full clinical abstraction of inpatient care.9,10 No inference is made in this presentation about the degree to which inpatient and postdischarge deaths, RApoLOS outliers, or 90-day readmissions are preventable. These outcomes are no-fault observations about current national performance in elective surgery for Medicare patients. However, from our prior observations with the National Inpatient
Sample, it is likely that these overall adverse outcomes of elective surgical care can be improved when a reference set of cost-effective hospitals is defined. Prediction models for readmission to the hospital have had poor discrimination.11 Better risk adjustment for all outcomes and costs of care through the full 90-day postdischarge period will provide goals for improvement. Furthermore, these data indicate that patients who are RApoLOS Outliers and even those with coded complications should receive special attention after discharge to reduce postdischarge adverse outcomes. Adverse outcomes and inefficiency are major drivers in the excess costs of US health care. It is unlikely that punitive government policies of nonpayment for non-risk-adjusted rates of readmissions, or nonpayment for ‘‘never events,’’ will have positive benefits for patients, nor will they reduce expenditures for care.12 With the rapid evolution of improved health care information technology and improved prediction modeling, it can be anticipated that full episodebased cost profiles that compute overall costs of care can be development for all elements of inpatient care inclusive of 90 days of postdischarge events. Risk-adjusted ‘‘bundled’’ payment for the whole episode will give providers the optimum incentive for best outcomes at the best cost.
References 1. Jencks SF, Williams MV, Coleman EA. Rehospitalizations among patients in the Medicare fee-for-service program. N Engl J Med 2009; 360:1418–28. 2. Centers for Medicare and Medicaid Services. Readmissions reduction program. Available at: http://www.cms.gov/Medicare/Medicare-Feefor-Service-Payment/AcuteInpatientPPS/Readmissions-ReductionProgram.html#. Accessed June 6, 2013.
330 3. Pine M, Fry DE, Jones BL, et al. Screening algorithms to assess the accuracy of present-on-admission coding. Perspect Health Inf Manag 2009;6:2. 4. Fry DE, Pine M, Jones BL, et al. Surgical warranties to improve quality and efficiency in elective colon surgery. Arch Surg 2010;145: 647–52. 5. Pine M, Fry DE, Jones BL, et al. Controlling costs without compromising quality: paying hospitals for total knee replacement. Med Care 2010;48:862–8. 6. Fry DE, Pine M, Jones BL, et al. Adverse outcomes in surgery: redefinition of post-operative complications. Am J Surg 2009;197: 479–84. 7. Fry DE, Pine M, Jones BL, et al. Control charts to identify adverse outcomes in elective colon resection. Am J Surg 2012;203:392–6. 8. Wheeler DJ. Understanding variation: the key to managing chaos. 2nd ed. Knoxville, TN: SPC Press; 2000. p. 137. 9. Pine M, Jordan HS, Elixhauser A, et al. Enhancement of claims data to improve risk adjustment of hospital mortality. JAMA 2007;297: 71–6. 10. Fry DE, Pine M, Jordan HS, et al. Combining administrative and clinical data to stratify surgical risk. Ann Surg 2007;246:875–85. 11. Kansagara D, Englander H, Salanitro A, et al. Risk prediction models for hospital readmission: a systematic review. JAMA 2011;306: 1688–98. 12. Fry DE, Pine M, Jones BL, et al. Patient characteristics and the occurrence of never events. Arch Surg 2010;145:148–51.
Discussion Darrell A. Campbell, Jr, M.D. (Ann Arbor, MI): This is important from my perspective as a chief medical officer at the University of Michigan. Recently we slipped a couple notches in our US News and World Report rankings. And so guess who gets the attention about that. It’s me. Don has really been a leader in trying to help us go through all that data from 4,000 hospitals and actually figure out something that means good qualitydrisk adjusted post-op length of stay, RApoLOS. And now he’s showing up this morning that it’s also associated with post discharge deaths and readmissions, so that’s a reflection of quality also. But there is a few, you know, gaps in the knowledge at this point, one of which is that you talked about this a little bit, that there are a fair number of patients who don’t have any coded complications, but they’re 3s RApoLOS, patients, and those who didn’t have a SNF to go to is socioeconomic thing. What about the readmissions part it as a reflection of quality, if you say that 90 percent of
The American Journal of Surgery, Vol 207, No 3, March 2014 them are related to the index hospitalization, I will believe you, but at our hospital when we’ve looked at that, it hasn’t been that high. Is this ready for prime time now as a metric for quality that we could all use? And if you don’t think so, then what are the steps that we need to do to get it there? Donald E. Fry, M.D. (Chicago, IL): I would say that if we looked at emergency operations, your issue about disposition of the patient would have some validity, however, these were elective operations in every case. They were elective operations that were operated on within 48 hours of admission, and I find it unacceptable to believe that 16% of the outliers would relate to disposition problems if the patient was electively being put in the hospital. Obviously, what we need is a clinical database to look at those cases and identify them. Is the model ready for prime time? I tend to believe that it is. I wish I had more clinical data to enhance the quality of the risk model, but the data looks pretty striking when you actually map it out and look at it, and so I would make the argument that if the patient survives the hospitalization, survives 90 days post discharge, was not a risk adjusted length of stay outlier and wasn’t readmitted within 90 days, that’s a pretty good place to start. James R. DeBord, M.D. (Peoria, IL): Don, you have showed us that it is more dangerous to be discharged from the hospital than to be admitted. Was there any correlation in readmission whether the patient was discharged to a care facility versus home? And do you believe that if postoperative care was, in fact, reimbursed, that surgeons would see these patients more frequently and that some of these readmissions could be reduced? Dr Fry: Because the majority of the readmissions actually in the Jencks study were medical readmissions, not surgical readmissions, having a financial incentive to see the patients didn’t seem to encourage them being seen more often. The SNF versus home is a very difficult issue because SNF admissions, long-term admissions are very much an optional thing in terms of the family situation and in terms of the availability of those services. So we have trieddwe are now in our next iteration looking at socioeconomic issues, SNF, the patient’s income, even, to see whether those things actually predict readmissions.
Midwest Surgical Association
Medicare post-discharge deaths and readmissions following elective surgery Donald E. Fry, M.D.a,b,c,*, Michael Pine, M.D, M.B.A.a,d, Gregory Pine, B.A.a a
Michael Pine & Associates, 1 East Wacker Drive, #1201, Chicago, IL 60601, USA; bDepartment of Surgery, Northwestern University Feinberg School of Medicine, Chicago, IL, USA; cDepartment of Surgery, University of New Mexico School of Medicine, Albuquerque, NM, USA; dDepartment of Medicine, The University of Chicago, Chicago, IL, USA
KEYWORDS: Risk-adjusted outcomes; Elective surgical care; Hospital readmissions; Postdischarge death rates; Control charts; Postoperative mortality rates
Abstract BACKGROUND: The frequency of 90-day, postdischarge deaths and readmissions in Medicare patients undergoing elective surgical procedures has not been well studied. METHODS: The Medicare MedPar database for 2009 to 2010 was used to develop inpatient risk-adjusted, postoperative length-of-stay (RApoLOS) prediction models for live discharges in 21 categories of elective operations. Moving average control charts were used in each category to define RApoLOS outliers (.3s). The relationships between RApoLOS outliers and all postdischarge deaths and readmissions within 90 days of discharge were assessed. RESULTS: The inpatient mortality rate was .5%. Of 2,054,189 live discharges, 147,292 (7%) were RApoLOS outliers. There were 14,657 postdischarge deaths (.7%) and 187,566 readmissions (9%). RApoLOS outliers had a 3.5% death rate and a 17% rate of readmission, while those found not to be RApoLOS outliers had a .5% death rate and a 9% readmission rate (P , .0001). CONCLUSIONS: RApoLOS outliers have increased rates of postdischarge deaths and readmissions. Ó 2014 Elsevier Inc. All rights reserved.
Complications of inpatient care of medical and surgical patients are identified as a cause of deaths, patient morbidity, and excess costs to the health care system. Many reports identify deaths and complications that occur only during the actual inpatient phase of patient management, and postdischarge events are not reported in large part because of the difficulty in getting accurate postdischarge data. Recent data indicate that as many as one third of hospitalized Medicare patients are readmitted to the hospital ,90 days after an
The authors declare no conflicts of interest. * Corresponding author. Tel.: 11-312-467-1700; fax: 11-312-4671705. E-mail address: [email protected] Manuscript received July 14, 2013; revised manuscript September 1, 2013 0002-9610/$ - see front matter Ó 2014 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.amjsurg.2013.09.007
inpatient episode of care.1 Many of these readmissions follow emergency medical and surgical hospitalizations. A plausible argument can be made that postdischarge readmission of Medicare patients may be driven by the progression of underlying disease risk factors and that these readmissions may not be avoidable. Others contend that readmissions are a direct consequence of the index hospitalization and should be viewed as complications of care. All agree that readmissions are expensive to the health care system. With major inpatient elective operations, postdischarge complications are a source of concern, but the actual frequency of these events has been difficult to define. Postdischarge events are not systematically reported, and even after elective procedures, patients may be readmitted to hospitals other than the sites of their index operations. The costs of readmission have reached a significant level of
D.E. Fry et al.
Postdischarge deaths and readmissions
concern, resulting in the implementation of financial penalties by the Centers for Medicare and Medicaid Services for those rates of readmissions that are deemed to be excessive.2 In this study, we evaluated postdischarge deaths and readmissions to acute care hospitals among a large population of Medicare patients undergoing a broad array of different elective surgical procedures. We tested the hypothesis that major adverse outcomes of the index hospitalization are a predictor of 90-day postdischarge death and readmission.
Methods We used the MedPar database from Medicare for 2009 and 2010 to evaluate the postdischarge events of patients undergoing 21 different categories of elective surgical procedures. The categories were 2 neurosurgery; 1 face, head, and neck; 1 lung and thorax; 2 cardiac, 3 vascular; 2 gastrointestinal; 1 miscellaneous high risk; 1 laparoscopic cholecystectomy; 1 hepatobiliary and pancreatic; 1 abdominal wall hernia; 1 renal and hysterectomy; 1 prostate and urinary; 1 female reproductive (not hysterectomy); 1 breast; 1 vertebral surgery; and 1 hip and knee replacement. Multiple categories were defined in neurosurgery, cardiac surgery, vascular surgery, and gastrointestinal surgery to optimize uniformity in anticipated lengths of stay and death rates. For example, carotid endarterectomy has a very different length of stay and mortality rate compared with aortic aneurysm repair within vascular surgery. This Medicare data set permits the identification of administrative claims for all hospitalizations and permits the identification of all readmissions. It does not have Part B claims for physicians and ambulatory services. Dates of death for Medicare beneficiaries are identified and permit the determination of 90-day postdischarge mortality rates.
Defining the analytic database All cases within each surgical category were identified using International Classification of Diseases, Ninth Revision, codes. Only elective patients aged R65 years who underwent operations on days 0, 1, or 2 of hospitalization were included from hospitals with R20 cases for the study period. Inpatient deaths were identified and removed from the subsequent postoperative length-of-stay (poLOS) analysis. Present-on-admission quality screens were used to eliminate poor coding hospitals for model development, but these facilities were retained for all outcome analyses.3
Risk-adjusted poLOS outliers Previously we have identified risk-adjusted (RApoLOS) outliers to be patients with clinically relevant major complications after inpatient surgical operations.4,5 RApoLOS outliers refine the definition of a relevant complication of care in that they have .3s observed length of stay compared with predicted, and they have costs that are
327 statistically greater than patients with coded complications who are not RApoLOS outliers. To define these patients, a standard list of risk factors for clinical diagnoses present on admission for each of the 21 surgical groups was defined. A preliminary forward stepwise linear regression model for poLOS was then computed from those patients within each category who had no coded complications. Coefficients were determined for significant risk factors (P , .05). A unique prediction model was derived for each category of elective procedure. Hospital variables were used to eliminate these effects on coefficients of risk models. A moving average control chart by methods previously published was then created within each category of operation and within each hospital.6,7 The observed poLOS was identified, and the predicted poLOS was calculated from the specific prediction model. Within each hospital for each eligible procedure, the total predicted poLOS days were set equal to the total observed days by multiplication of predicted values by a constant. The observed-minuspredicted poLOS differences for all patients within each category were then determined in the temporal sequence within each hospital. Within each hospital, observedminus-predicted differences for each patient in the temporal sequence were subtracted from each other to give an absolute difference. The mean of all absolute differences within each hospital for each category was computed, and this mean was multiplied by a constant of 2.66 to define the 3s upper control limit of the control chart.8 All categories of procedures had outliers with dramatically extended poLOS. These dramatic outliers had a profound impact on the mean of absolute differences between sequential cases within a hospital. After the first control chart was completed, the outlier cases were removed, and the process was repeated with the revised mean of absolute differences to remove a second group, albeit smaller, number of outliers. The process was iteratively repeated until no outliers remained. In statistical process control terms, the remaining cases had only common-cause variation, and all remaining cases reflected an ‘‘in control’’ stable system. Total RApoLOS is the number of cases that needed to be removed to establish the stable system. All prediction models were then further refined by recalibration of the coefficients of the prediction model for each of the 21 categories, but this time only with patients who were not outliers. This enhanced the accuracy of overall models by removing the impact of outlier cases on final coefficients. This final RApoLOS model was then run against all nonoutlier cases from the preliminary evaluation to identify additional outliers after recalibration of coefficients. Rates of RApoLOS outliers were then computed for each of the 21 categories of operations.
Postdischarge deaths and readmissions All live discharges from the index hospitalization were then examined for 90 days after discharge. All 90-day
328
The American Journal of Surgery, Vol 207, No 3, March 2014
Table 1 Observed deaths and readmissions of all patients divided into those who were RApoLOS outliers and who had coded complications in the discharge abstract No RApoLOS outliers Variable No. of patients: (% of all patients) Postdischarge deaths (death rate) Odds ratio No. of readmissions (% of total patients) Odds ratio
RApoLOS outliers
No coded complications
Coded complications
No coded complications
Coded complications
Total patients
1,141,937 (55.6%)
764,960 (37.2%)
24,067 (1.2%)
123,225 (6.0%)
2,054,189
3,800 (.33%)
5,776 (.76%)
542 (2.28%)
4,557 (3.70%)
14,675 (.71%)
1.0 85,397 (7.48%)
2.3 77,592 (10.1%)
6.9 3,913 (16.3%)
11.5 20,664 (16.8%)
187,566 (9.13%)
1.0
1.4
2.4
2.5
RApoLOS 5 risk-adjusted, postoperative length of stay.
readmissions to acute care hospitals were identified for each category. It is appreciated that all readmissions may not be directly linked to the index hospitalization, especially in a Medicare population that may have a number of unrelated acute and chronic illnesses that may necessitate rehospitalization. For this initial analysis, we chose to include all readmissions because our preliminary observations led us to believe that 90% of readmissions within 90 days are linked to the index hospitalization. Death records were then queried to identify all live discharges who died during the 90 days after discharge. All computations and data analysis were performed using SAS version 9.1.3 (SAS Institute Inc, Cary, NC).
Results After all inpatient deaths from the index hospitalization were removed (.5%), a total of 2,054,189 live discharges were evaluated for postdischarge events. Within 90 days of discharge from the hospital, there were 14,675 deaths (.7%) and 187,566 readmissions (9%). Among live discharges, a total of 888,185 (43%) had R1 coded complications for the index hospitalization. A total of 147,292 patients (7%) were RApoLOS outliers. Table 1 lists the data relationships among RApoLOS outliers, identified coded complications, and 90-day postdischarge and readmissions. It should be emphasized that 37% of all discharges had coded complications but were not RApoLOS outliers. Also of note, 24,067 of all RApoLOS outliers (16%) did not have coded complications at discharge. Coded complications and RApoLOS were independently significant in association with postdischarge deaths and readmissions (P , .0001). However, patients with no coded complications who were RApoLOS outliers had a significantly higher rate of postdischarge death and readmission than did those patients with coded complications who were not RApoLOS outliers. Odds ratios predicting postdischarge death or readmission were highest among the patients who had coded complications and were RApoLOS outliers by univariate analysis (P , .0001).
RApoLOS outliers had a 3.5% postdischarge death rate and a 17% rate of readmission, while nonoutliers had a .5% death rate and a 9% readmission rate (P , .0001). Table 2 identifies rates of coded complications and RApoLOS outliers of specific high-volume operations. The 90-day postdischarge deaths and readmission rates for cardiac and colon surgery were higher than the mean for the entire patient population, while total joint replacement rates were lower.
Comments In this study, we examined elective surgical procedures that spanned a broad scope of severity and costs. The results indicate that more patients died in the 90 days after discharge than died as inpatients after the operations and that overall, 9% of these Medicare patients were readmitted to the hospital in this 90-day period. Not surprisingly, readmission rates were much higher for major cardiothoracic and abdominal operations. We have identified that RApoLOS outliers were statistically associated with 90-day postdischarge deaths and readmissions (P , .0001). These observations are consistent with our several publications across different elective surgical procedures that RApoLOS outliers represent significant inpatient morbidity and, on the basis of the current data, are associated with significant postdischarge adverse events. RApoLOS outliers who do not have any coded complications are of considerable interest. These outliers predict postdischarge deaths and readmissions with nearly the same statistical significance as the outliers with coded complications. The excess length of stay is unlikely to be the consequence of addition unnecessary days of hospitalization, because the Inpatient Prospective Payment System of the Centers for Medicare and Medicaid Services provides a negative incentive for extended hospitalization. It is likely that major complications of care have been omitted from the abstract summaries of these patients. In elective surgical care, the excess poLOS should seldom be driven by poor discharge planning or disposition issues. Administrative
D.E. Fry et al. Table 2 surgery
Postdischarge deaths and readmissions
329
Data on RApoLOS outliers, 90-day readmissions, and 90-day deaths that occurred in cardiac, joint replacement, and colon
Variable
Cardiac
Total joint replacement
Colon surgery
All patients
Inpatient mortality rate Total live discharges Index hospitalization No CC, no RApoLOS CC, no RApoLOS No CC, RApoLOS CC, RApoLOS Postdischarge deaths No CC, no RApoLOS CC, no RApoLOS RApoLOS Total readmissions No CC CC, no RApoLOS No CC, RApoLOS CC, RApoLOS
2.42% 155,596
.11% 682,459
2.72% 69,986
.54% 2,054,189
29,566 (19.0%) 108,052 (69.4%) 395 (.25%) 17,583 (11.3%) 1,929 (1.24%) 134 890 905 24,055 (15.5%) 3,672 16,996 69 3,318
327,434 (48.0%) 323,440 (47.4%) 3,671 (.54%) 27,914 (4.09%) 1,022 (.15%) 309 484 229 36,893 (5.41%) 14,868 18,553 289 3,183
23,065 (33.0%) 39,850 (56.9%) 238 (.34%) 6,833 (9.76%) 2,115 (3.02%) 277 1,326 512 10,018 (14.3%) 2,583 6,397 27 1,011
1,141,937 (55.6%) 764,960 (37.2%) 24,067 (1.2%) 123,225 (6.0%) 14,675 (.71%) 3,800 5,776 5,099 187,566 (9.13%) 85,397 77,592 3,913 20,664
CC 5 coded complications; RApoLOS 5 risk-adjusted, postoperative length of stay. Twenty-one categories of Medicare patients undergoing elective surgical procedures were evaluated. Risk-adjusted length-of-stay outliers from the index hospitalization predicted increased rates of 90-day postdischarge deaths and readmissions to the hospital.
issues in patient disposition would not explain the higher rates of deaths and readmissions after discharge. Another interesting population of patients is those with coded complications who are not RApoLOS outliers. These patients have higher readmission rates than those who do not have any coded complications. In the Medicare population, many of these coded complications may be driven by incentives for higher Medicare severity-diagnosisrelated group payments. The RApoLOS methodology does provide a measure of severity to the coded complications that can be of value in the identification of those patients in need of focused review of processes of care and those at greatest risk for readmission. There were 3 limitations of this study. First, all-cause deaths and readmissions are reported, and some may not be related to the index hospitalizations or operations. Interval trauma or unrelated diseases may account for a portion of the 90-day postdischarge events. Second, the absence of Part B data in this data set prevents a total cost analysis of the index and readmission hospitalizations. Finally, risk adjustment using administrative data is always challenged, and models with greater clinical data are always desirable. Accurate present-onadmission coding and the incorporation of readily available clinical elements such as admission laboratory data into the administrative data set have the prospect of enhancement of modeling without resorting to the more expensive process of full clinical abstraction of inpatient care.9,10 No inference is made in this presentation about the degree to which inpatient and postdischarge deaths, RApoLOS outliers, or 90-day readmissions are preventable. These outcomes are no-fault observations about current national performance in elective surgery for Medicare patients. However, from our prior observations with the National Inpatient
Sample, it is likely that these overall adverse outcomes of elective surgical care can be improved when a reference set of cost-effective hospitals is defined. Prediction models for readmission to the hospital have had poor discrimination.11 Better risk adjustment for all outcomes and costs of care through the full 90-day postdischarge period will provide goals for improvement. Furthermore, these data indicate that patients who are RApoLOS Outliers and even those with coded complications should receive special attention after discharge to reduce postdischarge adverse outcomes. Adverse outcomes and inefficiency are major drivers in the excess costs of US health care. It is unlikely that punitive government policies of nonpayment for non-risk-adjusted rates of readmissions, or nonpayment for ‘‘never events,’’ will have positive benefits for patients, nor will they reduce expenditures for care.12 With the rapid evolution of improved health care information technology and improved prediction modeling, it can be anticipated that full episodebased cost profiles that compute overall costs of care can be development for all elements of inpatient care inclusive of 90 days of postdischarge events. Risk-adjusted ‘‘bundled’’ payment for the whole episode will give providers the optimum incentive for best outcomes at the best cost.
References 1. Jencks SF, Williams MV, Coleman EA. Rehospitalizations among patients in the Medicare fee-for-service program. N Engl J Med 2009; 360:1418–28. 2. Centers for Medicare and Medicaid Services. Readmissions reduction program. Available at: http://www.cms.gov/Medicare/Medicare-Feefor-Service-Payment/AcuteInpatientPPS/Readmissions-ReductionProgram.html#. Accessed June 6, 2013.
330 3. Pine M, Fry DE, Jones BL, et al. Screening algorithms to assess the accuracy of present-on-admission coding. Perspect Health Inf Manag 2009;6:2. 4. Fry DE, Pine M, Jones BL, et al. Surgical warranties to improve quality and efficiency in elective colon surgery. Arch Surg 2010;145: 647–52. 5. Pine M, Fry DE, Jones BL, et al. Controlling costs without compromising quality: paying hospitals for total knee replacement. Med Care 2010;48:862–8. 6. Fry DE, Pine M, Jones BL, et al. Adverse outcomes in surgery: redefinition of post-operative complications. Am J Surg 2009;197: 479–84. 7. Fry DE, Pine M, Jones BL, et al. Control charts to identify adverse outcomes in elective colon resection. Am J Surg 2012;203:392–6. 8. Wheeler DJ. Understanding variation: the key to managing chaos. 2nd ed. Knoxville, TN: SPC Press; 2000. p. 137. 9. Pine M, Jordan HS, Elixhauser A, et al. Enhancement of claims data to improve risk adjustment of hospital mortality. JAMA 2007;297: 71–6. 10. Fry DE, Pine M, Jordan HS, et al. Combining administrative and clinical data to stratify surgical risk. Ann Surg 2007;246:875–85. 11. Kansagara D, Englander H, Salanitro A, et al. Risk prediction models for hospital readmission: a systematic review. JAMA 2011;306: 1688–98. 12. Fry DE, Pine M, Jones BL, et al. Patient characteristics and the occurrence of never events. Arch Surg 2010;145:148–51.
Discussion Darrell A. Campbell, Jr, M.D. (Ann Arbor, MI): This is important from my perspective as a chief medical officer at the University of Michigan. Recently we slipped a couple notches in our US News and World Report rankings. And so guess who gets the attention about that. It’s me. Don has really been a leader in trying to help us go through all that data from 4,000 hospitals and actually figure out something that means good qualitydrisk adjusted post-op length of stay, RApoLOS. And now he’s showing up this morning that it’s also associated with post discharge deaths and readmissions, so that’s a reflection of quality also. But there is a few, you know, gaps in the knowledge at this point, one of which is that you talked about this a little bit, that there are a fair number of patients who don’t have any coded complications, but they’re 3s RApoLOS, patients, and those who didn’t have a SNF to go to is socioeconomic thing. What about the readmissions part it as a reflection of quality, if you say that 90 percent of
The American Journal of Surgery, Vol 207, No 3, March 2014 them are related to the index hospitalization, I will believe you, but at our hospital when we’ve looked at that, it hasn’t been that high. Is this ready for prime time now as a metric for quality that we could all use? And if you don’t think so, then what are the steps that we need to do to get it there? Donald E. Fry, M.D. (Chicago, IL): I would say that if we looked at emergency operations, your issue about disposition of the patient would have some validity, however, these were elective operations in every case. They were elective operations that were operated on within 48 hours of admission, and I find it unacceptable to believe that 16% of the outliers would relate to disposition problems if the patient was electively being put in the hospital. Obviously, what we need is a clinical database to look at those cases and identify them. Is the model ready for prime time? I tend to believe that it is. I wish I had more clinical data to enhance the quality of the risk model, but the data looks pretty striking when you actually map it out and look at it, and so I would make the argument that if the patient survives the hospitalization, survives 90 days post discharge, was not a risk adjusted length of stay outlier and wasn’t readmitted within 90 days, that’s a pretty good place to start. James R. DeBord, M.D. (Peoria, IL): Don, you have showed us that it is more dangerous to be discharged from the hospital than to be admitted. Was there any correlation in readmission whether the patient was discharged to a care facility versus home? And do you believe that if postoperative care was, in fact, reimbursed, that surgeons would see these patients more frequently and that some of these readmissions could be reduced? Dr Fry: Because the majority of the readmissions actually in the Jencks study were medical readmissions, not surgical readmissions, having a financial incentive to see the patients didn’t seem to encourage them being seen more often. The SNF versus home is a very difficult issue because SNF admissions, long-term admissions are very much an optional thing in terms of the family situation and in terms of the availability of those services. So we have trieddwe are now in our next iteration looking at socioeconomic issues, SNF, the patient’s income, even, to see whether those things actually predict readmissions.
