Association between Hospital Safety-Net Status and Delivery of Rehabilitation to Older Adults with Acute Respiratory Failure

Abstract

Background:

Older adults with socioeconomic disadvantage suffer greater decline in function and cognition following critical illness, an adverse outcome potentially preventable through mobilization. Whether safety-net hospitals (SNHs) that serve the highest proportions of patients with socioeconomic disadvantage are less likely to deliver rehabilitation during hospitalization with stay in the intensive care unit (ICU) is unknown.

Research Question:

Are SNHs less likely to provide rehabilitation (physical and/or occupational therapy) services to older adults hospitalized with acute respiratory failure (ARF) who receive invasive mechanical ventilation (IMV) than non-SNHs?

Study Design and Methods:

A retrospective cohort study of older adults (age≥65 years) hospitalized with ARF, ICU stay≥1 day, and receipt of IMV between 2016–19 using Medicare Provider Analysis and Review files. The primary outcome was delivery of rehabilitation during hospitalization. The exposure was SNH status defined as hospitals in top quartile of Disproportionate Share Hospital index. We constructed hierarchical multivariable logistic regression models with hospitals as random effect, adjusting for patient and hospital characteristics, to evaluate the association between SNH status and rehabilitation delivery. We calculated hospital-level risk-standardized rehabilitation delivery rate and characterized variation using median odds ratio (MOR).

Results:

We identified 868,735 ICU hospitalizations across 1,859 US hospitals; half were adults between 65–74 years, 48% were female and 77% were of White race. Rehabilitation was delivered in 59.1% of all hospitalizations. In the adjusted model, SNHs had 20% lower odds of delivering rehabilitation compared with non-SNHs [Adjusted Odds Ratio (aOR) (95% CI): 0.80 (0.75–0.86)]. Hospitals varied widely in delivering rehabilitation services with a median risk-standardized rehabilitation rate of 59.4% (IQR 51.7, 67.3) and a MOR of 1.69.

Interpretation:

SNHs had 20% lower odds of delivering rehabilitation to older adults hospitalized with ARF; differences in rehabilitation delivery could be a potential mechanism for socioeconomic disparities in functional and cognitive decline after critical illness.

In the United States, over 1 million adults are admitted to intensive care units (ICUs) annually for acute respiratory failure (ARF) requiring invasive mechanical ventilation (IMV) ¹. Two-thirds survive the index hospitalization ¹, and up to 70% develop long-term functional and cognitive impairments ^2–4. Older adults are particularly vulnerable to decline in function and cognition after ICU hospitalization due to pre-existing risk factors such as frailty, disability, cognitive impairment, and multimorbidity ^5–7. However, even after accounting for these factors, older adults with socioeconomic disadvantage develop a 30% greater burden of disability and have 10-fold increased odds of developing dementia in the year after ICU hospitalization ⁸. Mobilization during critical illness, most commonly achieved through rehabilitation services such as physical and occupational therapy has been shown to improve functional independence at discharge and reduce long-term cognitive impairment ^9–11. Evaluation by rehabilitation services can also identify care needs such as assistive devices for post-ICU rehabilitation that can promote functional recovery ¹². Inequitable delivery of rehabilitation could therefore be a mechanism for disparities in ICU-related functional and cognitive decline.

In the US, safety-net hospitals (SNHs) provide medical care to patients regardless of insurance status or ability to pay¹³. Consequently, patients with poor socioeconomic status (e.g., low-income older adults) and those belonging to minoritized race or ethnicity groups that closely overlap with socioeconomic status are more likely to access SNHs than those from less vulnerable backgrounds¹⁴. Known to be financially constrained¹⁵, SNHs could be prone to lower rehabilitation delivery because of limited resources for staffing or investment in performance improvement activities such as standardized mobilization protocols - factors known to be associated with rehabilitation delivery ^16,17. SNHs serve the highest proportions of patients with socioeconomic disadvantage, therefore lower rates of rehabilitation delivery at these hospitals could underlie socioeconomic disparities in post-ICU functional and cognitive decline.

We sought to evaluate the association between SNH status and delivery of rehabilitation services to older adults hospitalized with ARF requiring IMV in a nationwide cohort of hospitals. We also investigated hospital-level variation in the delivery of rehabilitation services. We hypothesized that SNHs would be less likely to deliver rehabilitation compared to non-SNHs after accounting for pre- and in-hospitalization clinical needs and that hospitals would vary widely in rehabilitation delivery.

STUDY DESIGN AND METHODS

Data Sources and Study Population

We conducted a retrospective cohort study using claims records for inpatient hospitalizations from Medicare-certified hospitals for fee-for-service beneficiaries from 2016–19 in the Medicare Provider Analysis and Review (MEDPAR) Limited Data Set. We linked the data to American Hospital Association (AHA) annual survey files to derive hospital characteristics using hospital provider IDs and to the Centers for Medicare and Medicaid Services (CMS) hospital inpatient prospective payment system (IPPS) Impact files for corresponding fiscal years to determine hospital SNH status, as detailed below.

We included short-stay acute care hospitalizations of older adults (age ≥65 years) with a primary or secondary discharge diagnosis of ARF, ICU stay of ≥1 day, and receipt of IMV across the fifty US states and the District of Columbia. ICU stay was determined using revenue codes indicating admission to general, specialty, or coronary care units while excluding psychiatric and intermediate care units ¹⁸. The diagnosis of ARF and receipt of IMV were determined by the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) diagnosis, and procedure codes respectively (e-Table 1) ¹⁹. We selected a cohort of older adults who received IMV because the evidence to support rehabilitation is strongest in this population ²⁰. To improve the reliability of hospital-level estimates, we excluded hospitals with <2 ICU beds and those that contributed <25 observations to our sample annually ¹⁸. Figure 1 presents the assembly of our analytic sample. We followed reporting guidelines for observational studies (e-Table 2). The study was considered exempt by Yale University Institutional Review Board because of the use of deidentified data.

Figure 1. Consort Diagram for Assembly of the Analytic Sample.

From 2016 to 2019, 1,045,568 hospital stays from 2,842 unique hospitals met the inclusion criteria. Of these, 148,850 from 354 hospitals were excluded due to a small ICU size (<2 beds) and 25,113 from 609 hospitals were further excluded due to the low volume of observations meeting inclusion criteria per year (< 25). 2,870 observations from 20 hospitals were excluded because they could not be linked to the CMS IPPS Impact files for determination of safety-net status. The final analytic sample included 868,735 hospital stays from 1,859 unique hospitals. ARF=Acute Respiratory Failure, ICU=Intensive Care Unit, US=United States.

Outcome Ascertainment

The primary outcome was the delivery of rehabilitation services during ICU hospitalization. We ascertained the delivery of rehabilitation based on charges for physical and occupational therapy services in the MedPAR Limited Dataset. The charges are assigned on the basis of claims for revenue center codes 042x and 043x as validated in previous work ^21,22. Because of variation in cost-to-charge ratios by hospitals ²³, we ascertained rehabilitation delivery as binary yes/no for charges > 0 vs 0.

Exposure Ascertainment

Our primary exposure was SNH status. We defined SNHs as hospitals in the top quartile of the disproportionate share hospital (DSH) index derived from annual financial reports provided to CMS by hospitals for corresponding fiscal years as in prior studies ^24,25. The DSH index measures the proportion of low-income insured patients served by a hospital and is used by states to allocate Medicaid payments ¹³. It is calculated using the formula:

$$DSH=\frac{NumberofMedicareSupplementalSecurityIncomeDays}{TotalMedicareDays}+\frac{Medicaid,Non-MedicareDays}{TotalPatientDays}26.$$

Assessment of Covariates

We selected covariates a priori based on factors identified as relevant to rehabilitation delivery in prior literature and availability in our datasets ^17,27–30. We accounted for patient demographics - age (categorized into intervals to account for non-linearity) and sex; comorbidities determined on the basis of “present on admission (POA)” indicator on diagnosis codes and operationalized as the Elixhauser Comorbidity Index; severity of acute illness as determined by the count of organ dysfunction using a validated algorithm for administrative data (e-Table 1)³¹, and ICU type ascertained from ICU indicator codes. We described race and ethnicity, categorized as Asian, Black, Hispanic, North American Native, White, other, and unknown as reported in the Medicare data ³² but did not adjust for this in our main model to obtain effect estimates unattenuated by factors other than clinical need for rehabilitation. We included hospital characteristics shown to be associated with rehabilitation delivery in prior work - teaching affiliation ^17,29, profit status ²⁹, hospital size determined by number of hospital beds =^17,30, geographical region ²⁹, and location ^29,33.

Statistical Analysis

We described patient and hospital characteristics for ICU hospitalizations, overall, by delivery of rehabilitation, and by SNH status. We describe means with standard deviations (SD) or medians with interquartile ranges (IQR) and counts with proportions as appropriate. We constructed a hierarchical multivariable logistic regression model with delivery of rehabilitation services (yes/no) as the dependent variable and hospital SNH status (yes/no) as the primary independent variable of interest. We included a random intercept for hospital and adjusted for patient (individual-level) and hospital (group-level) characteristics as fixed effects as in prior work³⁴. From this model, we calculated the expected rate of rehabilitation delivery for each hospital based on the case mix of its patients. We derived the risk-standardized rate of rehabilitation delivery for each hospital as the ratio of the observed rate of the outcome to the expected rate multiplied by the unadjusted observed rate across the whole sample as in prior work developed for risk-adjustment in performance measures ^35–37. We calculated the median odds ratio (MOR) to evaluate variation across hospitals unexplained by patient and hospital characteristics. The MOR represents the change in odds of rehabilitation delivery at a higher- versus a lower-delivery hospital for a patient with identical demographic and clinical characteristics. The magnitude of the MOR can be directly compared to odds ratios for other covariates in the model to assess the contribution of hospitals towards differences in the rate of rehabilitation delivery^42.

To evaluate the robustness of the association between SNH status and rehabilitation delivery, we conducted several sensitivity analyses. First, given that patients with longer ICU and/or hospital stays could have greater eligibility periods to receive rehabilitation services, we added ICU and hospital lengths of stay (LOS) as covariates in separate models as in prior work evaluating rehabilitation delivery ^22,39. Second, because very short ICU stays (<2 days) for either a rapidly improving or worsening condition resulting in death (e.g., sudden pulmonary embolism) may not require or provide opportunity for rehabilitation, we repeated our analysis excluding these observations. Third, since the diagnosis codes for ARF and organ dysfunction were ascertained at the aggregate hospitalization level in our main analysis, there is a possibility that rehabilitation delivery could have temporally preceded these factors. To address this, we conducted a sensitivity analysis where the diagnosis codes for ARF and organ dysfunction were present on admission using the POA indicator in Medicare claims data. Fourth, while we did not adjust for race and ethnicity in the main model to derive estimates unattenuated by factors other than clinical care needs, we evaluated differences after accounting for it as a covariate in a sensitivity analysis. Fifth, we conducted a post-hoc interaction analysis to determine effect modification by hospital characteristics significantly associated with rehabilitation delivery. Finally, given variability in definitions of SNHs, we conducted a sensitivity analysis using an alternative definition wherein all public hospitals were considered SNHs⁴⁰. All analyses were performed using SAS (version 9.4; SAS Institute, Inc., Cary, NC). All tests were two-sided, and p-values <0.05 were considered statistically significant.

RESULTS

We identified 868,735 acute care hospitalizations of older adults with ARF who received IMV and had an ICU stay of at least 1 day across 1,859 unique hospitals (Figure 1). Patient and hospital characteristics of the study cohort are described and stratified by delivery of rehabilitation services in Table 1. Half of the hospitalizations were contributed by adults between 65–74 years of age (50.2%) and 48% by females.

Table 1. Characteristics of ICU hospitalizations of older adults with acute respiratory failure who received mechanical ventilation, overall and by delivery of rehabilitation.

ICU= intensive care unit, IQR=interquartile range, CI=confidence interval

	Overall	Rehabilitation Delivery
	(N=868,735)	Yes  (N=513,181)	No (N= 355,554)
Patient characteristics
Age, n (%)
65–69	237,987 (27.4)	145,095 (28.3)	92,892 (26.1)
70–74	197,805 (22.8)	122,405 (23.9)	75,400 (21.2)
75–79	173,282 (20.0)	104,744 (20.4)	68,538 (19.3)
80–84	128,037 (14.7)	73,452 (14.3)	54,585 (15.4)
85–89	85,908 (9.9)	46,142 (9.0)	39,766 (11.2)
90 and more	45,716 (5.3)	21,343 (4.2)	24,373 (6.9)
Sex, n (%)
Male	451,390 (52.0)	267,343 (52.1)	184,047 (51.8)
Female	417,345 (48.0)	245,838 (47.9)	171,507 (48.2)
a Race and Ethnicity, n (%)
Asian	16,701 (1.9)	8,488 (1.7)	8,213 (2.3)
Black	134,718 (15.5)	75,292 (14.7)	59,426 (16.7)
Hispanic	18,319 (2.1)	9,573 (1.9)	8,746 (2.5)
North American Native	5,289 (0.6)	3,131 (0.6)	2,158 (0.6)
White	670,666 (77.2)	403,352 (78.6)	267,314 (75.2)
Other	14,688 (1.7)	8,138 (1.6)	6,550 (1.8)
b ICU Type, n (%)
General	657,568 (75.7)	385,105 (75.0)	272,463 (76.6)
Medical	115,727 (13.3)	66,631 (13.0)	49,096 (13.8)
Surgical	64,295 (7.4)	41,911 (8.2)	22,384 (6.3)
Other	31,145 (3.6)	19,534 (3.8)	11,611 (3.3)
c ICU length of stay, days, median (IQR)	6 (3, 11)	8 (4,14)	3 (1,7)
c Hospital length of stay, days, median (IQR)	9 (4,16)	13 (8, 20)	4 (2, 9)
d Count of organ dysfunction, median (IQR)	2 (1, 3)	2 (1, 2)	2 (1, 3)
Elixhauser comorbidity index, mean (SD)	19.5 (16.8)	18.9 (16.8)	20.4 (16.8)
Hospital characteristics
e Safety-net status, n (%)	217,493 (25)	126,041 (24.6)	91,452 (25.7)
Teaching affiliation, n (%)	729,899 (84)	438,480 (85.4)	291,419 (82)
Profit status, n (%)
Not-for-profit	669,922 (77.1)	397,989 (77.6)	271,933 (76.5)
For-profit	97,714 (11.3)	55,764 (10.9)	41,950 (11.8)
Public	101,099 (11.7)	59,428 (11.6)	41,671 (11.7)
Hospital Beds, n (%)
<100	19,247 (2.2)	10,363 (2)	8,884 (2.5)
100–499	513,174 (59.1)	292,770 (57.1)	220,404 (62)
≥500	336,314 (38.7)	210,048 (40.9)	126,266 (35.5)
Census region, n (%)
South	370,751 (42.7)	212,838 (41.5)	157,913 (44.4)
Midwest	205,346 (23.6)	130,337 (25.4)	75,009 (21.1)
Northeast	159,620 (18.4)	94,611 (18.4)	65,009 (18.3)
West	133,018 (15.3)	75,395 (14.7)	57,623 (16.2)
Rurality, n (%)
Urban	826,726 (95.2)	490,521 (95.6)	336,205 (94.6)
Rural	42,009 (4.8)	22,660 (4.4)	19,349 (5.4)

^aRace and ethnicity are categorized as Asian, Black, Hispanic, White, North American Native and other as reported in Medicare data.

^bICU type (medical, surgical, general, or other) ascertained from ICU indicator codes in claims files, “other” category contains trauma, burn care, and other ICU types.

^cAscertained from hospitalization record in linked Medicare claims data.

^dCount of organ dysfunction during ICU hospitalization using an algorithm developed and validated for administrative data.

^eDefined as the top quartile of the disproportionate share hospital (DSH) index derived from annual Impact files provided to Centers for Medicaid and Medicare Services by hospitals for corresponding fiscal years.

Overall, rehabilitation was delivered in 59.1% of all hospitalizations. Hospitalizations during which rehabilitation was delivered were contributed by patients similar in age, sex, and severity of acute illness to those in which rehabilitation was not delivered (Table 1). A higher proportion of patients of Black race (16.7% vs 14.7%), greater burden of comorbidities [mean (SD): 20.4 (16.8) vs 18.9 (16.8)], and shorter ICU [median (IQR) 8 (4,14) vs 3 (1,7)] and hospital length of stay [13 (8,20) vs 4 (2,9)] was noted among hospitalizations during which rehabilitation was not delivered compared to those where it was delivered . Hospitals with a teaching affiliation (85.4 vs 82.0%) and large bed size (40.9% vs 35.5% for beds>500) were more likely to deliver rehabilitation than non-teaching and smaller hospitals, respectively. Table 2 presents patient and hospital characteristics by SNH status. Patients at SNHs were similar in demographics, comorbidities, severity of acute illness, and length of stay compared to those in non-SNHs. SNHs were more frequently at large hospitals (51.0% vs 34.6% hospitalizations) and those with teaching affiliations (91.5% vs 81.5% hospitalizations) than non-SNHs.

Table 2. Characteristics of ICU hospitalizations of older adults with acute respiratory failure who received mechanical ventilation, overall and by hospital safety-net status.

ICU= intensive care unit, IQR=interquartile range, CI=confidence interval

	Overall	Safety-net status
	(N=868,735)	Yes (N=217,493)	No (N= 651,242)
Patient characteristics
Age, n (%)
65–69	237,987 (27.4)	63,707 (29.3)	174,280 (26.8)
70–74	197,805 (22.8)	48,382 (22.3)	149,423 (22.9)
75–79	173,282 (20.0)	41,492 (19.1)	131,790 (20.2)
80–84	128,037 (14.7)	30,696 (14.1)	97,341 (15.0)
85–89	85,908 (9.9)	21,148 (9.7)	64,760 (9.9)
>= 90	45,716 (5.3)	12,068 (5.6)	33,648 (5.2)
Sex, n (%)
Male	451,390 (52.0)	114,256 (52.5)	337,134 (51.8)
Female	417,345 (48.0)	103,237 (47.5)	314,108 (48.2)
a Race and Ethnicity, n (%)
Asian	16,701 (1.9)	8,533 (3.2)	9,844 (1.5)
Black	134,718 (15.5)	49,896 (22.9)	84,822 (13)
Hispanic	18,319 (2.1)	10,518 4.8)	7,801 (1.2)
North American Native	5,289 (0.6)	1,752 (0.8)	3,537 (0.5)
White	670,666 (77.2)	141,289 (65)	529,377 (81.3)
Other	14,688 (1.7)	4,989 (2.3)	9,699 (1.5)
b ICU Type, n (%)
General	657,568 (75.7)	145,792 (67.0)	511,776 (78.6)
Medical	115,727 (13.3)	42,548 (19.6)	73,179 (11.2)
Surgical	64,295 (7.4)	21,243 (9.8)	43,052 (6.6)
Other	31,145 (3.6)	7,910 (3.6)	23,235 (3.6)
c ICU length of stay, days, median (IQR)	6 (3, 11)	6 (3, 12)	6 (3,11)
c Hospital length of stay, days, median (IQR)	9 (4,16)	10 (5,18)	9 (4,15)
d Count of organ dysfunction, median (IQR)	2 (1,3)	2 (1,3)	2 (1,3)
Elixhauser comorbidity index, mean (SD)	19.5 (16.8)	20.6 (17.1)	19.1 (16.7)
Hospital characteristics
Teaching affiliation, n (%)	729,899 (84.0)	198,950 (91.5)	530,949 (81.5)
Profit status, n (%)
Not-for-profit	669,922 (77.1)	134,105 (61.7)	535,817 (82.3)
For-profit	97,714 (11.3)	30,824 (14.2)	66,890 (10.3)
Public	101,099 (11.7)	52,564 (24.2)	48,535 (7.5)
Hospital Beds, n (%)
<100	19,247 (2.2)	1,307 (0.6)	17,940 (2.8)
100–499	513,174 (59.1)	105,172 (48.4)	408,002 (62.7)
≥500	336,314 (38.7)	111,014 (51.0)	225,300 (34.6)
Census region, n (%)
South	370,751 (42.7)	89,984 (41.4)	280,767 (43.1)
Midwest	205,346 (23.6)	33,219 (15.3)	172,127 (26.4)
Northeast	159,620 (18.4)	40,381 (18.6)	119,239 (18.3)
West	133,018 (15.3)	53,909 (24.8)	79,109 (12.2)
Rurality, n (%)
Urban	826,726 (95.2)	211,828 (97.4)	614,898 (94.4)
Rural	42,009 (4.8)	5,665 (2.6)	36,344 (5.6)

^aRace and ethnicity are categorized as Asian, Black, Hispanic, Non-Hispanic White, North American Native and other as reported in MedPAR database.

^bICU type (medical, surgical, general, or other) ascertained from ICU indicator codes in claims files, “other” category contains trauma, burn care, and other ICU types.

^cAscertained from hospitalization record in linked Medicare claims data.

^dCount of organ dysfunction during ICU hospitalization using an algorithm developed and validated for administrative data.

Figure 2 presents the distribution of median risk-standardized rehabilitation delivery rate (RSRR) across all hospitals in our cohort. Hospitals varied widely in the delivery of rehabilitation with a median RSRR of 59.4% (IQR: 51.7, 67.2). The MOR for hospitals was 1.69, meaning that for the same patient characteristics, a “high-rehabilitation delivery hospital” had nearly 70% greater odds of delivering rehabilitation during ICU hospitalization compared to a “low-rehabilitation delivery hospital” rate. In the fully adjusted model, SNHs had 20% lower odds of delivering rehabilitation services during hospitalization compared with non-SNHs [Adjusted Odds Ratio (aOR) (95% CI): 0.80 (0.75, 0.86)]. Rural hospitals had 11% lower odds of delivering rehabilitation than hospitals located in urban areas [aOR (95% CI):0.89 (0.82, 0.98)]. Large bed size (≥500 beds) [aOR (95% CI): 1.35 (1.19, 1.52)], location in the Midwest [aOR (95% CI):1.29 (1.19, 1.41)] and teaching affiliation [aOR (95% CI): 1.18 (1.11, 1.25)] were other hospital characteristics associated with higher rehabilitation delivery (Figure 3). In a post-hoc analysis evaluating the interaction between SNH status and rural location, the interaction term was not significant (P-value=0.32), suggesting no modification of the effect of SNH status on rehabilitation delivery by rurality.

Figure 2. Hospital Variation in Risk Standardized Rehabilitation Delivery Rate.

Hospital-level rates of rehabilitation were estimated from a hierarchical logistic regression model with rehabilitation delivery as the dependent variable and hospital as the random intercept, adjusting for patient and hospital characteristics listed as covariates in the Methods. The risk-standardized rehabilitation rate (RSRR) was calculated as the ratio of the predicted/ expected rate of rehabilitation delivery for each hospital multiplied by the unadjusted national observed rate as explained in the Methods. Q1= 1^st quartile, Q3 = 3^rd quartile.

Figure 3. Association Between Hospital- and Patient-level Characteristics and Rehabilitation Delivery during ICU Hospitalization.

Adjusted odds ratio (95% confidence intervals) for the outcome of rehabilitation delivered during ICU hospitalization was derived from a hierarchical multivariable logistic regression model with the hospital as a random intercept, adjusting for patient and hospital-level characteristics as fixed effects as explained in the Methods. aOR = Adjusted Odds Ratio, ICU= Intensive care unit, CI=confidence interval.

In sensitivity analysis adding race and ethnicity to the covariates in our model, the association between SNH and rehabilitation delivery and the MOR for hospital variation did not change substantially (e-Table 3). Compared with hospitalizations of patients with White race, those from other race and ethnicity groups had a reduced likelihood of rehabilitation delivery (e-Table 3). In sensitivity analyses adjusting for ICU and hospital LOS, the association between SNH status and rehabilitation delivery was stronger [aOR (95% CI):0.72 (0.67, 0.78) for ICU LOS and aOR (95% CI):0.65 (0.60, 0.71) for hospital LOS] (e-Tables 4 and e-Table 5). Hospital-level variation in rehabilitation delivery increased as evidenced by a higher MOR (1.87 after accounting for ICU LOS and 2.04 for hospital LOS). In sensitivity analyses including only observations with ICU LOS≥ 2 days and restricting diagnosis codes of ARF and organ dysfunction to POA the association between SNH status and rehabilitation delivery remained unchanged (e-Tables 6 and 7). Using an alternative definition of SNH identifying them as all public hospitals, the estimate was attenuated but remained similar in direction [aOR (95% CI):0.91 (0.84, 0.99) (e-Table 8)].

DISCUSSION

In this nationwide sample of Medicare-certified hospitals, we found that more than a third of older adults hospitalized with ARF who received IMV and had an ICU stay did not receive any rehabilitation services during hospitalization. SNHs, had 20% lower odds of delivering rehabilitation services during hospitalization than non-SNHs, and those in rural areas had 11% lower odds than those in urban areas. The treating hospital was the biggest contributor to variability in rehabilitation delivery, more than individual patient. Given the important role rehabilitation can play in mitigating functional and cognitive decline during ICU hospitalization ^9,11 and identifying needs for post-hospitalization recovery ¹², our findings warrant consideration of efforts to enhance equitable delivery of rehabilitation to critically ill older adults.

The prevalence of rehabilitation delivery to critically ill patients observed in our study is similar to that reported in clinician surveys³¹, point prevalence studies³², cohorts of clinical trial networks and quality improvement collaboratives^28,29,41 and reports during the COVID-19 pandemic ⁴². Despite growing evidence to support improvement in short- and long-term functional and cognitive outcomes with mobilization among critically ill adults ^11,20, our finding of rehabilitation delivery rates similar to that observed a decade ago ²⁹ highlights the need for implementation efforts to bridge the gap from evidence to care delivery. Our observation that more than a third of older adults did not receive rehabilitation services at any point during a hospitalization for ARF requiring IMV highlights a missed opportunity to reduce functional decline and identify post-acute rehabilitation needs.

To our knowledge, our study is the first to investigate structural differences at the hospital level in rehabilitation delivery. Since rehabilitation during hospitalization can mitigate functional and cognitive decline, differences in its delivery could serve as a mechanism for disparities in post-ICU impairments ^8,9,11. In recent work, we found that socioeconomic disadvantage and rural location were associated with a reduced likelihood of receiving rehabilitation services during an ICU hospitalization ⁴³. However, because of the small number of observations per hospital, we could not determine whether hospital-level effects were responsible for the observed differences. Our findings from the present study confirm that SNHs, are less likely to deliver rehabilitation services to critically ill older adults compared with non-SNHs. Although not the primary objective of our study, we also noted differences by race and ethnicity that deserve to be investigated in future work.

The significant variation at hospital-level in rehabilitation delivery and reduced delivery at SNHs and rural hospitals observed in our study could have several explanations. One possibility is reduced staffing by physical and occupational therapists who deliver rehabilitation or nurses who frequently identify the need and eligibility for it at under-resourced hospitals. In prior work, staffing was identified as an important determinant of rehabilitation delivery ^16,17,28. Another possible explanation is presence of mobilization protocols that have been demonstrated to improve rates of mobilization but require considerable institutional support ¹⁷. Since engagement in quality improvement collaboratives in critical care has largely been described among hospitals in urban areas ⁴⁴, the permeation of this knowledge into practice at hospitals in rural areas and those serving greater proportions of disadvantaged populations deserves to be investigated. SNHs are usually financially constrained¹⁵, therefore it is possible they lack the resources to ensure adequate staffing by rehabilitation therapists and invest in quality improvement activities such as mobilization protocols ^16,17.

In the US, insurers reimburse hospitals for inpatient rehabilitation services by fixed payments based on diagnosis bundles under the diagnosis-related group (DRG) payment system⁴⁵. In this system each admission is classified with a specific DRG and reimbursement accordingly. DRGs such as ARF does not include rehabilitation services and hospitals do not additional payment for delivering these services. Therefore, financially constrained hospitals may be less incentivized to deliver rehabilitation services in the absence of distinct compensation for critical illnesses such as ARF.

Our findings add to prior literature demonstrating socioeconomic disparities in the delivery of evidence-based critical care delivery. Uninsured patients are less likely to undergo invasive procedures such as tracheostomy and central venous access or receive renal replacement therapy during their ICU stay ⁴⁶ and low-income patients are less likely to receive lung protective ventilation ⁴⁷ and extracorporeal membrane oxygenation ⁴⁸ for ARF and circulatory support for cardiogenic shock ⁴⁹. That these differences existed in a universally insured population in our study suggests that factors beyond individual ability to pay and clinical needs are determinants of care delivery. This highlights the need to rigorously measure and understand differences in rehabilitation delivery that could underlie disparities in short- and long-term outcomes after critical illness.

Our results should be interpreted in the context of a few limitations. First, we could not distinguish the location of rehabilitation delivery - ICU versus wards, due to the lack of availability of this information in administrative data. Future studies distinguishing this could inform where efforts to promote rehabilitation delivery should be targeted. Second, we could not determine the timing and dose of rehabilitation services. While most evidence exists for the benefit of early mobilization in preventing ICU-associated functional and cognitive decline ²⁰, rehabilitation delivery later in the ICU or on wards could still improve long-term outcomes ⁵⁰. Third, we could not evaluate delivery of mobilization by nursing or other staff; however, in most US ICUs, evaluation by rehabilitation therapists is a part of mobilization protocols ¹⁶. Fourth, patient-level information on some factors that could affect rehabilitation delivery, such as pre-hospitalization care needs, use of vasopressors, renal replacement therapy, prone positioning, and extracorporeal circulatory support was not available in our data, making residual confounding a possibility. Nonetheless, the similarity of outcome rates and magnitude of hospital variation observed in our study to prior reports, adds validity to our findings ²⁹. Finally, different definitions of SNHs may capture varying hospital characteristics. In the absence of a consensus definition of SNHs, we utilized the most commonly used definition of SNHs as hospitals in the top quartile of DSH index²⁵. Our sensitivity analysis identifying SNHs as all public hospitals supports the robustness of our findings to alternative definition of SNHs ⁴⁰.

INTERPRETATION

In a national sample of older adults hospitalized with ARF who received IMV and had an ICU stay, we found that one-third did not receive any rehabilitation during hospitalization and that SNHs were less likely to deliver rehabilitation compared with non-SNHs. Our study highlights the need for implementation efforts to promote equitable delivery of rehabilitation during critical illness at financially constrained hospitals to potentially mitigate disparities in functional and cognitive decline after ICU hospitalization.

TAKE-HOME POINTS.

Study Question:

Are safety-net hospitals less likely to deliver rehabilitation services to older adults hospitalized with acute respiratory failure who receive invasive mechanical ventilation and have an ICU stay?

Results:

Over forty percent of older adults hospitalized with acute respiratory failure who received invasive mechanical ventilation did not receive any rehabilitation services during a hospitalization with ICU admission; safety-net hospitals and those in rural areas had lower odds of delivering rehabilitation services during an ICU hospitalization.

Interpretation:

Hospitals vary widely in the delivery of rehabilitation services to critically ill older adults with acute respiratory failure and safety-net hospitals are less likely to provide these services than non-safety-net hospitals.