1. The Effects of Critical Access Hospital Conversion on Patient Safety Pengxiang Li, PhD, University of Pennsylvania University of Iowa John E. Schneider, PhD, University of Iowa University of Iowa Marcia M. Ward, PhD, University of Iowa Support for this work was funded by the Agency for Healthcare Research and Quality through grant # HS015009

2. Background Many of the smallest rural hospitals were not able to recover their Medicare costs under the prospective payment system (PPS) rates (Dalton, et al, 2005; Stensland, et al, 2004) Medicare Rural Hospital Flexibility Program of the 1997 Balanced Budget Act  To protect small, financially vulnerable rural hospitals  Allowed hospitals meeting certain criteria to convert to critical access hospitals (CAH)  Changed Medicare reimbursement mechanism from prospective (PPS) to cost-based.  One of the objectives of the policy was to increase the quality of care in these hospitals

3. Patient Safety Definition  “freedom from accidental injury due to medical care, or medical errors.” (IOM, 1999)  “the failure of a planned action to be completed as intended or the use of a wrong plan to achieve an aim…[including] problems in practice, products, procedures, and systems.” (Quality Interagency Coordination Task Force, 2000) Each year, more than 44,000 Americans die in hospitals due to preventable medical errors (AHA, 1999).

4. Objective To examine the impact of critical access hospital (CAH) conversion on hospital patient safety in rural hospitals

5. Study Sample and Unit of Analysis Eight year panel data for 89 Iowa rural hospitals (rural PPS hospitals and CAHs) Unit of analysis is hospital-year Figure 1: Time Frame of Iowa Rural Hospitals CAH Conversion

6. Outcome Measurements AHRQ Patient Safety Indicators Created by AHRQ PSI SAS macro V3.0  Each observed rate of patient safety indicator can be defined as “the outcome of interest in the population at risk”  The risk-adjusted rate is the rate the provider would have if it had the same case-mix as the reference population (2003 state inpatient datasets from 38 states) given the provider’s actual performance.  AHRQ recommends suppressing the estimates if fewer than 30 cases are in the denominator Only five patient safety indicators are able to provide PSI measures for all rural Iowa hospitals  PSI-5: foreign body left during procedure  PSI-6: iatrogenic pneumothorax  PSI-7: selected infections due to medical care  PSI-15: accidental puncture or laceration  PSI-16: transfusion reaction Too rare to provide variability to differentiate hospitals in Iowa

7. Outcome Measurements AHRQ Patient Safety Indicators Four patient safety indicators (PSI-5, PSI-6, PSI-7 and PSI-15) are selected as the main measures for patient safety performance in Iowa rural hospitals. A composite patient safety indicator variable is the weighted average of each indicator  The weights are based on the frequency of the numerator of each PSI in our sample (8-year inpatient discharges among 89 Iowa rural hospitals). Binary PSIs  If the value of a PSI is higher than the median of the PSI in our sample, the binary variable for the PSI (poor performance) is equal to one; otherwise, it is equal to zero

8. CAH Variables CAH: CAH binary indicator  CAH it =1, if hospital i is in CAH status in year t  CAH it =0, if hospital i is in PPS status in year t CAHmv: moving average CAH variable  To examine the relatively long-term effects of CAH conversion  CAHmv it =(CAH it +CAH i(t-1) +CAH i(t-2) )/3  e.g. CAHmv=0.33 if the hospital is in the first year of conversion

9. Models Panel data tobit model  The distributions of PSIs are skewed  The value of PSIs is non-negative with a mass at 0  Tobit models have a conceptual advantage in analyzing data where the distribution of the dependent variable is normal above a limiting value  PSI it *= β 0 + β 1 CAH it + ∑β 2 X it + ∑β 3 Z t + ε it  PSI it = max (0, PSI it *)

10. Control variables X it : a vector of other explanatory variables for hospital i in year t  % Medicare patient days  % Medicaid days  The hospital mean of the Charlson comorbidity score  % surgical discharges  market concentration (HHI)  county-level per capita income  county-level population density Zt is a vector of the year dummy variables which will adjust the effects of unmeasured, time-specific factors

11. Models Problem with tobit model  Assumption: truncated normal distribution  A deviation to the assumption will lead to significant bias Using binary PSIs as dependent variables  Generalized Estimating Equations (GEE) Logit Model (predict odds of bad performance)  log(p it /1-p it )= b 0 + b 1 CAH it + ∑b n X it + ∑b m Z t + e it  The variance function is p it (1-p it )  We assume the within-subject association among repeated measures is a first-order autoregressive correlation pattern

12. Data Sources Iowa state inpatient datasets (SID) used to calculate PSIs for each hospital each year. Hospital case mix and HHI were calculated using Iowa SIDs. Other county variables were retrieved from Bureau of Health Professions Area Resource File (ARF)

13. Table 1: Means, Standard Deviation and Sources of Variables, 1997 and 2004 VariableData sources 19972004 MeanStd. Dev.MeanStd. Dev. PSI-5 (per 1000 discharges)Iowa SID, AHRQ0.040.220.070.33 PSI-6 (per 1000 discharges)Iowa SID, AHRQ0.471.370.180.81 PSI-7 (per 1000 discharges)Iowa SID, AHRQ0.831.810.270.90 PSI-15 (per 1000 discharges)Iowa SID, AHRQ2.443.221.783.24 Composite PSI (per 1000 discharges)Iowa SID, AHRQ3.784.462.303.65 CAHIHA000.740.44 CAHmvIHA000.620.43 % Medicare daysIowa SID66.68%11.33%65.71%14.88% % Medicaid daysIowa SID6.72%4.49%7.20%4.72% Hospital casemixCMS1.060.090.990.12 Market concentration (HHI)Iowa SID8,9361,9698,9252,008 County per capita incomeARF21,1801,84626,4014,338 County population densityARF31.8018.5131.8019.34

14. Table 2. Cross-sectional Comparison of Means of PSIs between Rural PPS and CAHs, 1997 to 2004 Year Hospital categories Number of HospitalsPSI-5PSI-6PSI-7PSI-15 Composite score of 4PSIs 1997Rural PPS890.040.470.832.441.93 1998Rural PPS890.050.270.362.722.04 1999Rural PPS880.020.30.692.992.29 CAH100000 2000Rural PPS780.060.360.63.092.35* CAH11000.412.121.58* 2001Rural PPS570.070.21**0.692.652.03 CAH3200.07**1.172.241.8 2002Rural PPS450.040.230.64**2.00**1.56** CAH440.210.340.14**1.89**1.41** 2003Rural PPS340.010.46**0.88**2.38**1.89** CAH550.060.26**0.41**1.89**1.45** 2004Rural PPS230.12*0.29**0.54**2.68**2.04** CAH660.06*0.14**0.17**1.46**1.09** * Significant difference in PSIs between rural PPS and CAH at p ≤ 0.10 (Wilcoxon rank sum test) **Significant difference in PSIs between rural PPS and CAH at p ≤ 0.05 (Wilcoxon rank sum test)

15. Figure 2. Changes in Patient Safety Indicators after Conversion

16. Table 3. GEE Logit Models of Binary PSIs (1=poor performance, 0=good performance) PSI-5PSI-6PSI-7PSI-15 Composite score of 4PSIs CAH-0.80-1.19**-1.26**-0.92**-0.70** % Medicare days -0.01 -0.01*-0.01 % Medicaid days -0.010.020.030.020.03 % of surgical discharges 0.09**0.05**0.010.18**0.16** Charlson Index 2.01**0.691.051.51**1.37** Market concentration (HHI) -2.110.370.24-0.21-0.15 Per capita income ($1,000) 0-0.030.090.060.07** Population density -0.010.020.0100 Intercept-3.23-2.67*-3.73**-3.87**-3.54** observations712 Note: Due to space limit, year dummy variables were omitted in this table. * Statistically significant at 0.10 level. **Statistically significant at 0.05 level.

17. Table 4. Sensitivity analysis: GEE models Models PSI-5PSI-6PSI-7PSI-15 Composite score of 4PSIs Models in Table 3 (89 hospitals, 1997 to 2004) CAH‡-0.8-1.19**-1.26**-0.92**-0.70** CAHmv-0.67-1.57**-2.05**-1.37**-1.03** Models adding proxy variable (the lag 1 year dependent variable) as covariates CAH-1.08-0.90**-1.13**-0.83**-0.64** CAHmv-1.06-1.08*-1.67**-1.08**-0.80** Models using DRG-weight as risk adjustment (89 hospitals, 1997 to 2004) CAH-0.78-1.19**-1.25**-0.92**-0.70** CAHmv-0.63-1.56**-2.07**-1.36**-1.03** Models excluding 8 hospitals which is in rural PPS in 2006 (81 hospitals, 1997 to 2004) CAH-0.94-1.11**-1.41**-0.85**-0.63** CAHmv-0.79-1.45**-2.28**-1.27**-0.94** Models adjusting for hospital transfer behavior ¥ CAH-0.84-1.24**-1.28**-0.93**-0.72** CAHmv-0.76-1.68**-2.12**-1.43**-1.12** * Statistically significant at 0.1 level. ** Statistically significant at 0.05 level. ‡ The same as the model in Table 3. ¥ Adding two variables (percentage of acute inpatient admission were transferred from other short-term hospital and percentage of acute inpatient patients were transferred to other short-term hospitals) into the models in Table II-6

18. Table 5. Sensitivity Analyses: Tobit Models of Continuous PSIs * Statistically significant at 0.1 level. ** Statistically significant at 0.05 level. ++ Convergence was not achieved. +++ Add hospital dummy variables in cross-sectional Tobit model ‡ The results should be interpreted with caution, given that estimations for some coefficients were not stable under quadchk ∏ PSI-5 are observed rate.

19. Results Cross-section and pre-post conversion comparisons showed that CAH hospitals had better performance of patient safety than rural PPS hospitals. The odds ratios of poor performance in CAH hospitals compared to rural PPS hospitals are 0.30 (CI: 0.14-0.64) for PSI-6, 0.29 (CI: 0.15- 0.56) for PSI-7, 0.40 (CI: 0.24-0.67) for PSI-15, and 0.49 (CI: 0.31- 0.80) for composite score of 4PSIs. CAH conversion had no significant impact on the observed rates of foreign body left during procedure. Moving average CAH indicator had larger effects than binary CAH scale. Sensitivity analyses using tobit models consistent results Findings were robust among sensitivity analyses using different samples and different methods

20. Better PSI Performance Associated CAH Conversion: Alternative Explanations Changes in coding behaviors  Did not find any significant change in hospital-level mean number of DX, mean Charlson score, mean number of Elixhauser comorbidities Changes in referral patterns (get less severe patients)  No significant change in percentage of patients transferred from other hospitals percentage of patients transferred to other hospitals  Models adjusting for referral behaviors show consistent results It reflects a trend toward improvement in patient safety for all hospitals  None-converted hospitals are comparison group  Add year dummy variables  Difference in difference

21. Conclusion and implication CAH conversion in rural hospitals resulted in enhanced performance of patient safety. After conversion, CAHs have better perform in patient safety in the long run. We speculate that the likely mechanism involved an increase in financial resources following CAH conversion to cost-based reimbursement for Medicare patients Limitations of the study  Administrative databases (missing codes, coding errors) We are not able to rule out the possibility of coding behavior change  The five indicators may not reflect the whole picture of patient safety in rural hospitals  CAH conversion variable is endogeneous.