Diagnosis-related group (DRG) based reimbursement creates incentives for reduction in hospital length of stay (LOS). Enhanced recovery and Perioperative Surgical Home programs achieve reductions in LOS, but often following multiple medical and social interventions, some with limited levels of evidence.1,2 We recently used the Nationwide Readmission Database to evaluate whether large hospitals in the United States with the briefest LOS for inpatient elective/scheduled surgery also had the greatest use of postacute care, specifically skilled nursing and/or inpatient rehabilitation facilities. Conceptually, hospitals could achieve briefer LOS through incentives and interventions resulting in the transfer of greater percentages of patients to postacute care facilities.3 Instead, while controlling for DRG, each 1-day decrease in hospital median LOS was associated with a significantly lesser odds of transfer to a postacute care facility (0.68; 95% confidence interval, 0.54–0.85; P= .0008).3
Our previous study was potentially important because postacute care is expensive, often lasting many days or weeks.4,5 However, the result was limited to the largest of US hospitals, 15 of the most common surgical DRGs, elective admissions, and data from a single year (2013). What is not known is the extent to which the findings are generalizable. In the current article, we study the same problem but using different years (2008 and 2014), survey design (US National Inpatient Sample), and statistical methodology. Our hypothesis was that we would confirm that reductions in hospital LOS were associated with significant reductions in the incidences of transfer to postacute care facilities. If our prior observation3 was robust to different included data sources and methods of analysis, senior hospital managers, perioperative medical directors, and anesthesiology department chairs could be more confident that they need not be concerned that promoting incremental (eg, 1-day) reductions in average LOS of many patients may unintentionally result in greater rates of transfers to short-term care facilities.5,6
Downloaded Summary Measures From HCUPnet
Summary measures were downloaded from the 2008 and 2014 US National Inpatient Sample via the Healthcare Cost and Utilization Project (HCUP) public portal, HCUPnet. The database was used subject to the Agency for Healthcare Research and Quality (AHRQ) data use agreement.7 The National Inpatient Sample provides US national estimates of summary measures by using a sample of discharges from all hospitals participating in HCUP.8
We selected National Inpatient Sample data from HCUPnet using 3 types of categories (ie, strata): (1) Clinical Classifications Software’s classes of procedures (hereafter abbreviated “CCS”),9 (2) DRGs including a major operating room procedure during hospitalization,10 or (3) CCS limiting patients to those with US Medicare as the primary payer. We used 3 different types of categories (strata) because each results in different adjustments for confounders. The 2 types of categories including CCS were restricted to operating room procedures, with the CCS studied being the principal procedure abstracted upon discharge. DRG controls for procedures, comorbidities, and complications.
The third type of category limited patients to US Medicare because, even with age adjustment, patients with Medicare as their primary insurance carrier have greater incidences of discharge to nursing home than do patients with commercial insurance.11,12 In the United States, patients without insurance have a lesser incidence of postacute care.12 Furthermore, when integrated over many hospital regions, the 2 principal factors influencing numbers of patients having surgery are the procedure (ie, DRG/CCS) and the payer.13,14 Between 2008 and 2014, most of the CCS we studied had an increase in the percentage of discharges with Medicare insurance.
For each category, and for each of the 2 years studied (ie, 2008 and 2014), summary measures downloaded from HCUPnet were the mean hospital LOS and incidence of discharge to a postacute care facility (skilled nursing or inpatient rehabilitation), including standard errors (SEs). Microsoft Excel 2010 Visual Basic for Applications (Redmond, WA) was used to make calls to HCUPnet and isolate desired portions of the produced web pages. Our data were not patients (eg, as would be listed in a STrengthening the Reporting of OBservational studies in Epidemiology [STROBE] checklist) but, rather, summary measures for the US national patient population.
For each of the 3 types of categories, we excluded categories (eg, CCS) that had (1) mean LOS in 2014 <2.5 days; (2) incidence of postacute care in 2014 was <5.0%; (3) SE of the change in LOS >0.50 days; and/or (4) category absent in either 2008 or 2014. The first criterion (ie, LOS <2.5 days) was used because the National Inpatient Sample only includes patients undergoing inpatient surgery. The categories with mean LOS of 1 or 2 days were for procedures often performed during outpatient surgery (ie, absent from the HCUPnet study dataset because there was no hospital admission). The second criterion was used because there needed to be a sufficiently large incidence of postacute care (ie, <5.0%) for there to be a substantive difference between years. For example, the DRG 766 cesarean delivery without complication or comorbidity had only 0.1% ± 0.03% of patients receiving postacute care.3 Even a doubling of the 0.1% incidence of postacute care would be of negligible quantitative importance but have a large weight in the linear regression because the SE was so small. The third criterion (ie, SE of LOS >0.5 days) had no effect on the results of the subsequent analysis (weighted linear regression), but reduced the categories in the figures, making them more interpretable.
As in our previous study, our inferential objective was to pool effects among many procedures (eg, multiple DRGs) so that our conclusions would be applicable to strategic decisions made by anesthesiology departments, because patients undergo many different procedures.3,15–18 For the current study, we tested the slope between (1) the pairwise change in mean LOS between 2008 and 2014 and (2) the pairwise change in the incidence of postacute care between 2008 and 2014. Complicating the linear regression, both variables (ie, pairwise changes) have large SE, as shown in Figures 1 to 3. For regular least squares linear regression, the independent variables are assumed to be measured without error.19–21 Using the bivariate weighted least squares regression, inverse weighting was based both on the squares of the SE of the change in mean LOS and change in mean incidence of postacute care. The bivariate weighted least squares regression was performed using the iterative method described by York and Williamson.19,20 The equations are summarized by Cantrell21 (equation 5). In the legends of each of the 3 figures, we include not only the results of the bivariate weighted linear regression but also, as sensitivity analyses, results for unweighted linear regression and for the Spearman rank correlation.
Greater reductions in the mean LOS between 2008 and 2014 were associated with smaller percentages of patients with disposition to postacute care (all 9 slopes P < .0001). Analyzed by CCS, DRG, or CCS limited to Medicare patients, each pairwise reduction in the mean LOS by 1 day was associated with an estimated 2.6% ± 0.4%, 2.3% ± 0.3%, or 2.4% ± 0.3% (absolute) pairwise reduction in the mean incidence of use of postacute care, respectively (Figures 1–3). We repeated the analyses using the 44 CCS limited to patients 45 to 64 years or 64 CCS limited to patients 65 to 84 years, and the slopes were 2.9% ± 0.5% and 2.3% ± 0.3%, respectively. We also repeated the analysis using the most common International Classifications of Diseases, Ninth Revision, Clinical Modification code within each CCS22,23; among the 44 usable codes, the slope was 4.5% ± 0.9%. Based on the finding of a positive slope with each of the 12 different methods of analysis, including several different patient populations and procedures, the hypothesis was accepted that reductions in hospital LOS were associated with reductions in the incidences of transfer to postacute care facilities and vice versa.
Enhanced recovery and perioperative surgical home programs achieve reductions in LOS, but often following multiple interventions, some with limited levels of evidence.1,2 Senior hospital managers, perioperative medical directors, and chairs of anesthesiology departments often lack quantitative knowledge of how individual changes made by these programs will influence end points other than hospital LOS. In addition, anesthesiology departments care for patients undergoing a wide diversity of procedures.3,15–18 Our previous study had patient-specific data (eg, comorbidities), but considered only 15 common DRGs, elective admissions, and large hospitals.3 In the current study, we were able to study 72 different CCSs and 174 DRGs over 2 different years. Still, just as we concluded from our previous study,3 averaged over many surgical procedures and hospitals, leaders can be confident that it is unlikely that reductions in LOS cause premature discharging of surgical patients to postacute care facilities, despite incentives produced for this by the DRG based reimbursement system.
For multiple reasons, we recommend being conservative and limiting the conclusion drawn from our results to be that reductions in LOS are not causing greater use of postacute care nationwide. First, the results are heterogeneous among procedures (Figures 1–3). We have deliberately described the overall effect averaged over many different procedures and specialties. We did so with the intent for our result to be useful from the perspective of anesthesiologists, physicians at postacute care facilities, insurers, and hospital administrators, because these individuals are responsible for patients undergoing many procedures.3,15–18 Second, the magnitude of the reduction (≅2.6% per 1-day reduction in mean LOS) is small. Achieving a reduction in the mean LOS of 1 day nationwide takes technological change (eg, new surgical approach or enhanced recovery protocols). Third, although our study design and conclusions were not impacted by the etiologies of changes in LOS and/or postacute care, we do not know the mechanism(s) responsible for our findings. In our previous study, we obtained the same result with different data and design, and neither individual patient’s age, comorbidities, or LOS changed the associations.3 From the current data, while controlling the false discovery rate at α = .05, there are 2 individual CCS with significant reductions in both the LOS and the incidence of use of postacute care (Figure 1). These are CCS 153 “hip replacement” and CCS 152 “arthroplasty knee.” These are the 2 surgical procedures with widespread use of bundled payments in the United States, including postoperative care.24–26 Therefore, clinicians likely are trying to reduce both the LOS and the use of postacute care, not one in lieu of the other. Our results show that efforts at reducing one does not systematically result in increases in the other nationally and when averaged among multiple procedures, despite financial incentives for this to occur.
Figures 1 to 3 probably show increases in mean LOS for multiple categories of procedures. This finding has been reported previously for Medicare patients between 2004 and 2014.27 Increases in the percentage of cases that are outpatient can be associated with increased LOS among the remaining patients. Our limiting consideration to categories of procedures with mean LOS of at least 2.5 days may represent a limitation. US Medicare pays for skilled nursing facilities postoperatively only after an LOS of 3 days.28
In conclusion, in the United States, nationally, averaged over many surgical procedures, reductions in hospital LOS were not associated with greater incidences of use of postacute care. Rather, for those procedures with reductions in hospital LOS over time, the use of postacute care was slightly reduced, as found in our previous study with a different national survey, different years, and different methodology.3
Name: Franklin Dexter, MD, PhD.
Contribution: This author helped design and conduct the study, analyze the data, and write the manuscript.
Name: Richard H. Epstein, MD.
Contribution: This author helped design the study and write the manuscript.
This manuscript was handled by: Nancy Borkowski, DBA CPA, FACHE, FHFMA.
1. Thompson EG, Gower ST, Beilby DS, et al. Enhanced recovery after surgery program for elective abdominal surgery at three Victorian hospitals. Anaesth Intensive Care. 2012;40:450–459.
2. Dexter F, Wachtel RE. Strategies for net cost reductions with the expanded role and expertise of anesthesiologists in the perioperative surgical home. Anesth Analg. 2014;118:1062–1071.
3. Dexter F, Epstein RH, Dexter EU, Lubarsky DA, Sun EC. Hospitals with briefer than average lengths of stays for common surgical procedures do not have greater odds of either re-admission or use of short-term care facilities. Anaesth Intensive Care. 2017;45:210–219.
4. Stowers MD, Lemanu DP, Hill AG. Health economics in Enhanced Recovery After Surgery programs. Can J Anaesth. 2015;62:219–230.
5. Crosby G, Culley DJ, Dexter F. Cognitive outcome of surgery: is there no place like home? Anesth Analg. 2014;118:898–900.
6. Cyriac J, Garson L, Schwarzkopf R, et al. Total joint replacement perioperative surgical home program: 2-year follow-up. Anesth Analg. 2016;123:51–62.
11. Lim HJ, Hoffmann R, Brasel K. Factors influencing discharge location after hospitalization resulting from a traumatic fall among older persons. J Trauma. 2007;63:902–907.
12. Sacks GD, Hill C, Rogers SO Jr. Insurance status and hospital discharge disposition after trauma: inequities in access to postacute care. J Trauma. 2011;71:1011–1015.
13. Dexter F, Wachtel RE, Sohn MW, Ledolter J, Dexter EU, Macario A. Quantifying effect of a hospital’s caseload for a surgical specialty on that of another hospital using multi-attribute market segments. Health Care Manag Sci. 2005;8:121–131.
14. Dexter F. Factors substantively influencing numbers of surgical cases performed at a research hospital. J Res Hosp. 2017;2:6.
15. Dexter F, Wachtel RE, Yue JC. Use of discharge abstract databases to differentiate among pediatric hospitals based on operative procedures: surgery in infants and young children in the state of Iowa. Anesthesiology. 2003;99:480–487.
16. Dexter F, Ledolter J, Hindman BJ. Quantifying the diversity and similarity of surgical procedures among hospitals and anesthesia providers. Anesth Analg. 2016;122:251–263.
17. Dexter F, Epstein RH, Dutton RP, et al. Diversity and similarity of anesthesia procedures in the United States during and among regular work hours, evenings, and weekends. Anesth Analg. 2016;123:1567–1573.
18. Dexter F, Epstein RH, Sun EC, Lubarsky DA, Dexter EU. Readmissions to different hospitals after common surgical procedures and consequences for implementation of perioperative surgical home programs. Anesth Analg. 2017;125:943–951.
19. York D. Least squares fitting of a straight line with correlated errors. Earth Planet Sci Lett. 1969;5:320–324.
20. Williamson JH. Least-squares fitting of a straight line. Can J Phys. 1968;46:1845–1847.
21. Cantrell CA. Review of methods for linear least-squares fitting of data and application to atmospheric chemistry problems. Atmos Chem Phys. 2008;8:5477–5487.
22. O’Neill L, Dexter F, Park SH, Epstein RH. Uncommon combinations of ICD10-PCS or ICD-9-CM operative procedure codes account for most inpatient surgery at half of Texas hospitals. J Clin Anesth. 2017;41:65–70.