Reducing Readmissions for Chronic Obstructive Pulmonary Disease in Response to the Hospital Readmissions Reduction Program

Abstract

Rationale: In August 2013, the Hospital Readmission Reduction Program announced financial penalties on hospitals with higher than expected risk-adjusted 30-day readmission rates for Medicare beneficiaries hospitalized for chronic obstructive pulmonary disease (COPD) exacerbation. In October 2014, penalties were imposed. We hypothesized that penalties would be associated with decreased readmissions after COPD hospitalizations.

Objectives: To determine whether the announcement and enactment of financial penalties for COPD were associated with decreases in hospital readmissions for COPD.

Methods: We used data from California’s Office of Statewide Health Planning and Development to examine unplanned 30-day all-cause and COPD-related readmissions after COPD hospitalization. The preannouncement period was January 2010 to July 2013. The postannouncement period was August 2013 to September 2014. The postenactment period was October 2014 to December 2017. Using interrupted time series, we investigated the immediate change after the intervention (level change) and differences in the preintervention and postintervention trends (slope change).

Results: We identified 333,429 index hospitalizations for COPD from 449 California hospitals. Overall, 69% of patients had Medicare insurance. For all-cause readmissions, the level change at announcement was 0.16% (95% confidence interval [CI], −1.07 to 1.38; P = 0.80); the change in slope between preannouncement and postannouncement periods was −0.01% (95% CI, −0.15 to 0.13; P = 0.92). The level change at enactment was 0.29% (95% CI, −1.11 to 1.69; P = 0.68); the change in slope between postannouncement and postenactment was 0.04% (95% CI, −0.10 to 0.18; P = 0.57). For patients with COPD-related readmissions, the level change at the time of the announcement was 0.09% (95% CI, −0.68 to 0.85; P = 0.83); the change in slope was 0.003% (95% CI, −0.08 to 0.09; P = 0.94). The level change at the time of the enactment was 0.22% (95% CI, −0.69 to 1.12; P = 0.64); the change in slope was −0.02% (95% CI, −0.10 to 0.07; P = 0.72).

Conclusions: We did not detect decreases in either all-cause or COPD-related readmission rates at either time point. Although this would suggest that the Hospital Readmission Reduction Program penalty was ineffective for COPD, COPD readmissions had decreased at an earlier time point (October 2012) when penalties were announced for conditions other than COPD. Based on this, we believe early, broad interventions decreased readmissions, such that no difference was seen at this later time points despite institution of COPD-specific penalties.

Over 16 million Americans have chronic obstructive pulmonary disease (COPD) (1). Although hospitalizations are necessary in <10% of patients with COPD exacerbations, they disproportionately account for 65% of the healthcare costs associated with COPD (2–4). One-fifth of patients hospitalized for COPD exacerbation have a readmission within 30 days (5, 6). Furthermore, among Medicare patients, COPD is the third leading cause of readmissions (7).

In an effort to curb readmission rates for the most common conditions, the Centers for Medicare and Medicaid Services (CMS) enacted the Hospital Readmissions Reduction Program (HRRP) in 2010. This program penalized hospitals for excess 30-day, all-cause, risk-standardized readmission rates for certain target conditions (i.e., congestive heart failure, myocardial infarction, and pneumonia). COPD became a target condition in October 2014.

Our previous multistate analysis demonstrated that enactment of the HRRP penalties for conditions other than COPD was associated with a decrease in both all-cause and COPD-related readmissions (8). In the current study, we focused on the later time point (October 2014) when COPD became a target condition. Previous studies have examined the time point when penalties were enacted for target conditions other than COPD (9, 10). Determining the effectiveness of the policy for COPD is important given 1) the implication of this measure on quality of care and 2) the financial implications of the program on hospitals. Using California state data, we hypothesized that patients hospitalized for COPD after it became a target condition would have lower all-cause 30-day readmission rates compared with the expected rate of readmission based on underlying trends.

Methods

Study Design and Data Source

We performed an interrupted time series analysis using administrative data from the California Office of Statewide Health Planning and Development (OSHPD) between the years 2010 and 2017 (11). These data contain all hospital discharges within the state of California. We chose California because 1) it has the largest population of any state, 2) it is diverse; 3) there are relatively few border states, which is important because we would likely capture the majority of readmission encounters; 4) data are available that span 2010 and 2017, which covers the entire period from announcement of the HRRP to more than 1 year after enactment of penalties for COPD; 5) encounters for individuals are linked across years; 6) data are granular enough to pinpoint the exact month of the policy change; and 7) the hospital variables are more complete in OSHPD data than the Statewide Inpatient Database, which is important because the HRRP financial penalty is enacted at the hospital level (12). OSHPD data have been used for rigorous health services research (13). In addition, at the time of the current analysis, not all states from our previous study using Statewide Inpatient Database (8) had 2017 and 2018 data available.

The project was exempted by the Partners Institutional Review Board (2020P001684).

Patient Sample

We studied adults aged ≥40 years, who were admitted to the hospital for COPD exacerbation. This age threshold has been used in previous studies to exclude patients with asthma exacerbations (14). The International Classification of Diseases, Ninth Revision (ICD-9) and International Classification of Diseases, Tenth Revision (ICD-10), Clinical Modification codes used by the CMS were used to define COPD hospitalizations and COPD-related readmissions (Table E1 in the online supplement).

Inclusion and Exclusion Criteria

Inclusion and exclusion criteria were similar to those of Myers and colleagues (8) and will be briefly described here. Patients who died during the index hospitalization, left against medical advice, or were transferred between hospitals were excluded. Only the first hospitalization within 30 days was considered a readmission. Patients admitted after 30 days were eligible to contribute a second index hospitalization based on the initial inclusion and exclusion criteria. Planned readmissions in the 30-day postdischarge period were excluded according to the CMS algorithm (15). Death in the 30 days after discharge was not treated as a competing risk per CMS methodology. In accordance with previous analyses, patients admitted in December 2017 were excluded because they could not contribute 30 days of follow-up (16).

Key Variables

Data elements included patient- and hospital-level characteristics (17). Patient-level variables include age, sex, race/ethnicity, primary health insurance, hospital length of stay, number of comorbidities from CMS (18), and discharge quarter. Hospital-level variables included urbanicity, teaching hospital, number of beds, ownership, safety-net status, and Case Mix Index. Small and rural hospital designation was determined by Section 124,840 of the California Health and Safety Code. Safety-net hospitals were defined as those in which the percentage of Medicaid and uninsured discharges fell in the top quartile for that year. The Case Mix Index is the average relative diagnosis-related group weight of a hospital’s inpatient discharges (19). A higher Case Mix Index (>1.0) indicates a more complex and resource-intensive patient load.

Key Dates

The interrupted time series had two time points of interest, namely, the announcement of penalties for COPD as a target condition in August 2013 and the enactment of penalties for COPD as a target condition in October 2014. Using these time points, we created three distinct time periods, as shown in Table E2. These include 1) before the announcement of penalties for COPD as a target condition (January 2010–July 2013), 2) after the announcement of penalties for COPD as a target condition but before the penalties were actually enacted (August 2013–September 2014), and 3) after the enactment of penalties for COPD as a target condition (October 2014–December 2017).

Exposure

The primary exposure was HRRP’s enactment of financial penalties for COPD as a target condition in October 2014. A previous study had examined the association between an earlier time point (i.e., the enactment of the financial penalties for target conditions other than COPD) (8).

Outcomes

The primary outcome was the level change in the interrupted time series for unplanned, all-cause, 30-day readmissions in October 2014. This is the difference between the expected and actual readmission rates when the penalties were enacted for COPD as a target condition. The expected readmission rate was estimated by the linear trend in the period preceding October 2014. We also examined the difference in slopes between the preenactment and postenactment periods. As a secondary outcome, we examined the earlier time point (August 2013) when the penalties were announced for COPD as a target condition, including both the level change and change in slopes. We also examined COPD-related readmissions because it is the most common cause of readmission after hospitalizations for COPD (20).

Statistical Analysis

We calculated monthly hospital-level, risk-standardized readmission rates using hierarchical logistic regression models following previously described methodology used by the CMS (18, 21). Separate models were run for all-cause and COPD-related readmissions. All models included a hospital-specific random intercept to handle clustering, were adjusted for the comorbidities described by the CMS, and included indicators for month and year. After fitting each model, we estimated the predicted and expected number of readmissions for each hospital. The ratio of predicted to expected for each hospital was then multiplied by the overall observed readmission rate for that month to obtain the hospital-specific risk-standardized rate.

We then calculated the statewide monthly mean adjusted readmission rates for the interrupted time series analysis. We used the hospital-level risk-standardized readmission rates following CMS methodology to better simulate what would likely have resulted in hospital-level responses to the penalty. We fit ordinary least squares regression models with Newey-West standard errors (22, 23) to handle heteroskedasticity. The presence of autocorrelation was assessed using the Ljung-Box test (24), and an appropriate lag term was included (up to 12 mo).

To differentiate the contribution of patient- and hospital-level factors in the hierarchical models, we report the variability of the random effect and intraclass correlation in models including the intercept alone (no patient factors) and with patient factors for all-cause and COPD-related readmissions. The intraclass correlation can be interpreted as the proportion of variance in the outcome explained by clustering within hospitals.

Analyses were performed using SAS 9.4 (SAS Institute) and STATA IC 15.1 (StataCorp). Two-tailed P < 0.05 was the threshold for statistical significance.

Sensitivity Analyses

As a sensitivity analysis, we restricted the sample to only Medicare beneficiaries given that the penalty was based on the readmission rates of Medicare patients. It was possible that hospitals targeted the Medicare population for their interventions. We also performed the interrupted time series with only the later time point (enactment in October 2014) to have a longer lead-in period, which we will call the preenactment period. The enactment time point was chosen for the sensitivity analysis versus the earlier announcement time period because we believed that the incentive to act would have been greater when the financial penalties were active.

Results

Of 406,887 hospitalizations for COPD, we identified 333,429 index hospitalizations from 449 hospitals between 2010 and 2017. Baseline characteristics of patients during the index stays are shown by year in Table 1. The median age was 70 years (interquartile range [IQR] 61–80). Approximately 57% of patients were female, 64% were non-Hispanic White, and 69% had Medicare insurance. The majority (56%) of hospitalizations lasted for 2–4 days, and 87% of patients had more than one comorbidity. We report the characteristics of the subgroup of patients who were readmitted in Table E3. Overall, the patient characteristics during readmission stays were similar to the characteristics of index stays.

Table 1.

Characteristics of patients with index hospitalizations for chronic obstructive pulmonary disease, 2010–2017 (n = 333,429)

Patient variables	Overall	2010	2011	2012	2013	2014	2015	2016	2017
Age, median (IQR), yr	70 (61–80)	70 (60–80)	70 (60–80)	70 (60–80)	71 (61–80)	70 (60–79)	70 (61–80)	70 (61–80)	71 (62–81)
Age, 40–64 yr	114,597 (34)	17,230 (35)	16,307 (35)	15,548 (35)	14,504 (34)	13,919 (36)	13,147 (34)	11,742 (34)	12,200 (31)
Age, ≥65 yr	218,832 (66)	31,710 (65)	30,597 (65)	28,690 (65)	28,055 (66)	24,353 (64)	25,016 (66)	22,774 (66)	27,637 (69)
Sex, n (%)
F	190,359 (57)	28,438 (58)	27,083 (58)	25,429 (57)	24,386 (57)	21,968 (57)	21,770 (57)	19,393 (56)	21,892 (55)
M	143,069 (43)	20,502 (42)	19,821 (42)	18,809 (43)	18,173 (43)	16,304 (43)	16,393 (43)	15,122 (44)	17,945 (45)
Race/ethnicity
Non-Hispanic White	213,438 (64)	32,156 (66)	30,336 (65)	28,366 (64)	27,132 (64)	24,122 (63)	23,746 (62)	22,020 (64)	25,560 (64)
Non-Hispanic Black	43,168 (13)	6,118 (13)	5,922 (13)	5,790 (13)	5,405 (13)	5,286 (14)	5,240 (14)	4,659 (14)	4,748 (12)
Hispanic	44,033 (13)	6,217 (13)	6,067 (13)	5,963 (13)	5,817 (14)	5,201 (14)	5,195 (14)	4,359 (13)	5,214 (13)
Other	32,790 (10)	4,449 (9)	4,579 (10)	4,119 (9)	4,205 (10)	3,663 (10)	3,982 (10)	3,478 (10)	4,315 (11)
Primary health insurance
Medicare	230,882 (69)	33,532 (69)	32,493 (69)	30,409 (69)	29,602 (70)	25,962 (68)	26,390 (69)	23,936 (69)	28,558 (72)
Medicaid	58,967 (18)	7,790 (16)	7,324 (16)	6,934 (16)	6,654 (16)	7,868 (21)	7,678 (20)	7,242 (21)	7,477 (19)
Private	30,520 (9)	5,075 (10)	4,603 (10)	4,311 (10)	3,857 (9)	3,472 (9)	3,362 (9)	2,740 (8)	3,100 (8)
Other	13,060 (4)	2,543 (5)	2,484 (5)	2,584 (6)	2,446 (6)	970 (3)	733 (2)	598 (2)	702 (2)
Hospital length of stay
0–1 d	41,641 (12)	5,859 (12)	5,698 (12)	5,566 (13)	5,355 (13)	4,828 (13)	4,787 (13)	4,408 (13)	5,140 (13)
2–4 d	187,912 (56)	26,620 (54)	25,813 (55)	24,958 (56)	24,059 (57)	22,373 (58)	21,926 (57)	19,787 (57)	22,376 (56)
>4 d	103,876 (31)	16,461 (34)	15,393 (33)	13,714 (31)	13,145 (31)	11,071 (29)	11,450 (30)	10,321 (30)	12,321 (31)
Number of CMS comorbidities
0–1 comorbidity	43,525 (13)	6,413 (13)	6,068 (13)	5,909 (13)	5,870 (14)	5,350 (14)	5,200 (14)	3,749 (11)	4,966 (12)
2–3 comorbidities	129,431 (39)	15,963 (33)	15,482 (33)	15,112 (34)	14,910 (35)	13,980 (37)	15,403 (40)	17,153 (50)	21,428 (54)
≥4 comorbidities	160,473 (48)	26,564 (54)	25,354 (54)	23,217 (52)	21,779 (51)	18,942 (49)	17,560 (46)	13,614 (39)	13,443 (34)

Definition of abbreviations: CMS = Centers for Medicare and Medicaid Services; IQR = interquartile range.

Data are presented as n (%) unless otherwise specified.

The number of total hospitalizations, index hospitalizations, and readmissions remained relatively constant over time. The range of index hospitalizations was 34,516–48,940 per year (Figure E1). In addition, the risk-adjusted readmission rate for all-cause and COPD-related readmissions was relatively constant over time (mean, 18.3% and 6.5%, respectively; Figure E2). The range of all-cause readmission rate was 16.9–19.5%; the range for COPD-related readmission rate was 5.2–7.4%. Figure E2 also confirms that the percentage of all-cause and COPD-related readmissions specifically within the Medicare subpopulation remained relatively stable over time (mean, 18.6% and 6.0%, respectively). In the Medicare subgroup, the range for all-cause readmission rate was 17.1–19.9%; the range for COPD-related readmission rate was 4.7–7.0% (Figure E2).

Characteristics of hospitals are shown in Table 2. There was very little turnover in hospitals over time (11 hospitals did not have data spanning the full 8-yr period). Most hospitals were urban (86%), nonteaching hospitals (92%) without safety-net status (76%). The median number of hospital beds was 50 beds (IQR, 68–273). Lastly, the median Case Mix Index was 1.3 (IQR, 1.1–1.5). When we estimated the adjusted readmission rates, the intraclass correlation coefficient at the hospital level was approximately 1.3%, indicating that very little of the variation was due to the clustering by hospital.

Table 2.

Characteristics of hospitals, 2010–2017 (n = 449)

Hospital variables	n	Percentage or IQR
Urbanicity*
Urban	386	86
Rural (including small)	63	14
Teaching hospital
No	415	92
Yes	34	8
Number of beds
Median	50	68–273
0–99 beds	156	35
100–199 beds	115	26
200–299 beds	73	16
300–399 beds	43	10
≥400 beds	62	14
Safety-net hospital†
No	341	76
Yes	108	24
CMI‡
Median	1.3	1.1–1.5
CMI >1.00	403	90

Definition of abbreviations: CMI = Case Mix Index; IQR = interquartile range.

Data are presented as n (%) unless otherwise specified. Eleven hospitals were missing data for ≥1 year.

^*Urban designation was determined by Section 124,840 of the California Health and Safety Code.

^†Safety-net hospitals were defined as those in which the percentage of Medicaid and uninsured discharges fell in the top quartile for that year.

^‡The CMI is the average relative diagnosis-related group weight of a hospital’s inpatient discharges. A higher CMI (>1.0) indicates a more complex and resource-intensive case load.

We present the interrupted time series analysis for all-cause and COPD-related readmission, which we show for all patients as well as for only the Medicare subgroup (Figures 1 and 2, respectively). For patients with any insurance who were readmitted for any reason, the level change at the time of the announcement was 0.16% (95% confidence interval [CI], −1.07 to 1.38; P = 0.80). This can be interpreted as the mean increase in readmission rate across hospitals at the time of the announcement. The change in slope from the preannouncement period to postannouncement period was −0.01% (95% CI, −0.15 to 0.13; P = 0.92). This can be interpreted as the slope decreasing by 0.01% between the preannouncement and postannouncement period. The level change at the time of the enactment was 0.29% (95% CI, −1.11 to 1.69; P = 0.68); the change in slope from the postannouncement period to the postenactment period was 0.04% (95% CI, −0.10 to 0.18; P = 0.57; Figure 1A).

Figure 1.

Interrupted time series analysis of patients readmitted within 30 days for any reason, including (A) all patients and (B) Medicare subgroup. Each dot represents the statewide monthly mean adjusted readmission rate. There are two critical time points labeled. We derived a line of best fit for each of three segments, accounting for autocorrelation with 1-month and 5-month lag terms, respectively. Patients could be readmitted with any diagnosis. (A) For all patients, the level change at the time of the announcement was 0.16% (95% confidence interval [CI], −1.07 to 1.38; P = 0.80); the change in slope was −0.01% (95% CI, −0.15 to 0.13; P = 0.92). The level change at the time of penalty enactment was 0.29% (95% CI, −1.11 to 1.69; P = 0.68); the change in slope was 0.04% (95% CI, −0.10 to 0.18; P = 0.57). (B) For Medicare patients, the level change at the time of announcement was −0.37% (95% CI, −1.25 to 0.51; P = 0.41); the change in slope was 0.02% (95% CI, −0.10 to 0.13; P = 0.78). The level change at the time of penalty enactment was −0.12% (95% CI, −1.37 to 1.13; P = 0.85); the change in slope was 0.02% (95% CI, −0.10 to 0.13; P = 0.78).

Figure 2.

Interrupted time series of patients readmitted within 30 days for COPD, including (A) all patients and (B) Medicare subgroup. Each dot represents the statewide monthly mean adjusted readmission rate. There are two critical time points labeled. We derived a line of best fit for each of three segments, accounting for autocorrelation with 1-month lag terms. Patients were readmitted with chronic obstructive pulmonary disease. (A) For all patients, the level change at the time of the announcement was 0.09% (95% confidence interval [CI], −0.68 to 0.85; P = 0.83); the change in slope was 0.003% (95% CI, −0.08 to 0.09; P = 0.94). The level change at the time of penalty enactment was 0.22% (95% CI, −0.69 to 1.12; P = 0.64); the change in slope was −0.02% (95% CI, −0.10 to 0.07; P = 0.72). (B) For Medicare patients, the level change at the time of announcement was −0.31% (95% CI, −1.00 to 0.38; P = 0.38); the change in slope was 0.03% (95% CI, −0.06 to 0.13; P = 0.51). The level change at the time of penalty enactment was −0.05% (95% CI, −1.05 to 0.95; P = 0.92); the change in slope was −0.05% (95% CI, −0.14 to 0.04; P = 0.30). COPD = chronic obstructive pulmonary disease.

For the Medicare subgroup, the level change at the time of announcement was −0.37% (95% CI, −1.25 to 0.51; P = 0.41); the change in slope from the preannouncement period to the postannouncement period was 0.02% (95% CI, −0.10 to 0.13; P = 0.78). The level change at the time of penalty enactment was −0.12% (95% CI, −1.37 to 1.13l P = 0.85); the change in slope from the postannouncement period to the postenactment period was 0.02% (95% CI, −0.10 to 0.13; P = 0.78; Figure 1B). For all of these models, CIs for the β coefficient crossed zero. In other words, there does not appear to be a statistically significant difference in the level changes or slope changes for all-cause readmissions at either time point for the cohort with any insurance or for the Medicare subgroup.

For patients with any insurance whose readmissions were COPD related, the level change at the time of the announcement was 0.09% (95% CI, −0.68 to 0.85; P = 0.83); the change in slope from the preannouncement period to the postannouncement period was 0.003% (95% CI, −0.08 to 0.09; P = 0.94). The level change at the time of penalty enactment was 0.22% (95% CI, −0.69 to 1.12; P = 0.64); the change in slope from the postannouncement period to the postenactment period was −0.02% (95% CI, −0.10 to 0.07; P = 0.72; Figure 2A).

For the Medicare subgroup, the level change at the time of announcement was −0.31% (95% CI, −1.00 to 0.38; P = 0.38); the change in slope from the preannouncement period to the postannouncement period was 0.03% (95% CI, −0.06 to 0.13; P = 0.51). The level change at the time of penalty enactment was −0.05% (95% CI, −1.05 to 0.95; P = 0.92); the change in slope from the postannouncement period to the postenactment period was −0.05% (95% CI, −0.14 to 0.04; P = 0.30; Figure 2B). Similar to above, there does not appear to be a statistically significant difference in the level changes or slope changes for COPD-related readmissions at either time point for the cohort with any insurance or for the Medicare subgroup.

Finally, when we performed the interrupted time series analysis with only one time point (enactment of penalties in October 2014), one model became statistically significant. For all-cause readmissions in the Medicare subgroup, the change in slope was statistically significant (0.04%; 95% CI, 0.01–0.06; P = 0.003). This can be interpreted as a relatively small but statistically significant increase in the trend of COPD-related readmissions for Medicare patients from before penalties were enacted for COPD (preenactment period) to after penalties were enacted for COPD (postenactment period). There were no differences in the level changes or slopes changes for all-cause readmissions for patients with any insurance (Figure E3), nor were there differences for COPD-related readmissions for patients with any insurance or in the Medicare subgroup (Figure E4).

Table E4 shows the contribution of patient- and hospital-level factors in the hierarchical regression models for all-cause and COPD-related readmission for all patients and only Medicare patients. Model 1 includes the intercept only (no patient factors); model 2 contains patient factors. There is little difference in variability of the random effect between models 1 and 2 for both outcomes and patient populations, indicating a relatively small contribution of patient factors. The intraclass correlations were also relatively small for both outcomes and patient populations, indicating a low level of clustering within hospitals.

Discussion

We found that hospital-level rates of nonelective, 30-day, all-cause and COPD-related readmissions did not decrease either at the time of announcement (August 2013) or at the time of enactment (October 2014) of financial penalties for COPD as a target condition in California hospitals. We analyzed both the level changes and the changes in slope before and after the time points of interest. Our results are internally consistent across models; in the model of all-cause readmissions in the Medicare subgroup in which only one time point was examined, there was a statistically significant increase in the slope, not a decrease. Our rates for both all-cause (17–20%) and COPD-related (5–7%) readmissions are consistent with the literature (20, 25).

Given the negative main result, our findings call into question how effective the HRRP penalties are for COPD, especially given the sizable amount of money at risk for hospitals (i.e., strong financial incentive). An alternative explanation for the negative result in this paper might be that off-target effects of the HRRP occurred for COPD before COPD became a target condition. A previous study has shown that COPD readmissions decreased at an earlier time point (October 2012) when financial penalties were enacted for target conditions other than COPD (8). Indeed, in that previous study, both all-cause and COPD-related readmissions decreased for patients with any insurance type (8). The authors argue that broad interventions to prevent readmissions took place with the first wave of penalties; there may be a limit to the extent that readmissions can be reduced even if COPD later becomes a target condition.

A recently published study by Puebla Neira and colleagues (26) reported findings that conflict with this study. However, the data source, designation of time windows, and comparisons were different. Puebla Neira and colleagues defined their time windows around a blackout period in their data, which occurred because of the merging of two datasets that resulted in incomplete data for the year 2012. They compared the first half of the preannouncement period, which they defined as December 2006 to July 2008, with the last half of the implementation period, which they defined as May 2016 to November 2017. They reported a decrease in 30-day, all-cause readmissions as well as an association between higher mortality and HRRP implementation (26). We attribute the difference in findings for readmission to the aforementioned differences; Puebla Neira and colleagues (26) defined time windows around a blackout period and compared a very late segment to a very early segment in time, whereas we defined our time windows around two prespecified time periods and compared the level changes and changes in slopes surrounding the two prespecified time points.

The finding from Puebla Neira and colleagues (26) of increased mortality associated with the HRRP has been challenged by Lindenauer and colleagues (27), who demonstrated a shift of patients with primary diagnosis of pneumonia to a primary diagnosis of COPD (and a secondary diagnosis of pneumonia) starting in October 2016. They report that patients with a secondary diagnosis of pneumonia had higher 30-day mortality (∼11%) compared with patients without a secondary diagnosis of pneumonia (∼6%) (27), potentially causing a false association between mortality and the HRRP for COPD. Other studies have shown an association between worse mortality and the HRRP for other conditions, such as heart failure and pneumonia, so the association seems plausible (10). However, notably, if there were an association between HRRP and mortality such that patients died before being readmitted, it would have been more likely for us to detect an association between HRRP and decreased readmissions, but we did not. The current study does not include postdischarge mortality data. Because the HRRP does not count death as a competing risk when penalties are calculated, we did not.

Several studies have examined the efficacy of interventions to reduce COPD readmissions (5, 28–30). Results have been mixed. Many of the studies have involved an intervention with a care bundle. Parikh and colleagues (31) implemented an electronic care bundle for COPD. The 30-day readmission rate was statistically lower in patients who received the intervention (9% vs. 54%). They also showed a decrease in hospital-related costs in the intervention group (31). Zafar and colleagues. (32) implemented a multicomponent care bundle consisting of 1) an appropriate inhaler regimen, 2) a 30-day inhaler supply, 3) inhaler education, 4) a follow-up appointment within 15 days, and 5) standardized patient-centered postdischarge instructions. They achieved >90% adherence to the bundle over a 5.5-month period of iteration and decreased readmissions from 22.7% to 14.7%. In addition, receiving more of the bundle was associated with decreased probability of 30-day readmission (32). Meanwhile, Aboumatar and colleagues (33) performed a randomized controlled trial of a 3-month program that combined transitional support and long-term self-management support. They reported that patients in the intervention arm had more COPD-related hospitalizations/emergency department visits without a significant change in quality of life. It is unclear why patients experienced more events in this trial. Many of these interventions did not take into account patient input or target mental health or sleep health, even though patients with COPD have comorbid depression, anxiety, and sleep apnea. In addition, social factors may be under addressed, including tobacco dependence, need for sustainable exercise regimens, and access to portable oxygen devices.

Given these previous studies, we evaluated the contribution of patient- and hospital-level factors to our hierarchical models to potentially inform interventions to prevent COPD readmissions. Both of these groups of factors contributed only modestly to the models, which suggests that other factors are important. These could be COPD-specific severity factors (such as oxygen dependence or spirometry) or social factors (such as smoking status or inhaler adherence). For the current study, we simulated the CMS risk adjustment method by including patient- and hospital-level factors that are available in administrative data and used by the HRRP to enact penalties. Future research is needed to understand what factors, especially modifiable factors, can be the crucial levers we use to decrease readmissions for COPD.

The results should be interpreted in the context of the study design. First, the sensitivity of administrative codes to identify patients with COPD can be as low as 25% (34), and administrative data do not contain COPD-specific variables, such as spirometry (forced expiratory volume), to use in the risk adjustment. Because we used the same diagnosis codes and risk adjustment schema as the CMS and the HRRP, the patients represented in this study should at least be the patients on whom the penalty is based. Second, this California dataset contained fewer hospitals compared with our previous analysis of 904 hospitals (8), but notably the number of hospitalizations in this study was higher than those of other studies published on COPD readmissions (6). In addition, we used data from a large, diverse state that linked patients over the 8-year time period and contained granular enough data to report readmission rates at the monthly level. As previously mentioned, all states used in our prior multistate publication (8) did not offer the later years of data, which were critical to create a robust postenactment period slope. This was a necessary aspect to answer the question of interest.

The study has several strengths. The analysis was performed at the hospital level, which simulates the HRRP penalty structure. This is different from previous studies on COPD readmissions that have been performed at the patient level (8, 20). We used patient-level predictors to derive the hospital-level expected readmission rates. In addition, the risk adjustment technique also simulated the HRRP methodology, as previously stated.

In summary, we intended to study the hospital-level impact of the enactment of financial penalties for COPD under the HRRP. We hypothesized that there would be a decrease in 30-day all-cause readmissions for COPD in October 2014 because financial penalties enacted at that time would incentivize hospitals to alter their behavior. However, we did not find this despite performing several sensitivity analyses.

Conclusions

We did not detect a decrease in all-cause and COPD-related readmissions in October 2014 after COPD became a target condition under the HRRP. This was true regardless of patients’ insurance type and whether the preannouncement period was examined using one or two time points. Further research is needed to better understand how the HRRP changed behavior over time and whether there were unintended consequences.