The Impact of Personal and Historical Factors That Contribute to Medicare Readmissions

Abstract

Medicare readmissions from the 5% national sample of Medicare Fee For Service claims were assessed to identify the contribution of various demographic or personal health history risk factors to the liklihood of readmission within 30 days of discharge. The Medicare population was evaluated as 2 cohorts based on their eligibility status: age (71.9%) or disability (18.1%). Roughly 12% of admissions for the aged population resulted in a readmission within 30 days, whereas ∼18% was true of the disabled population. Patients with a history of frequent emergency department visits or an urgent index admission had an increased risk for a readmission in both groups of beneficiaries. An important indicator for readmission risk was discharge status from the index hospital stay. In both categories of beneficiaries, individuals who were discharged against medical advice were 1.6 times more likely to experience a readmission. Most importantly, personal and social determinants of health (SDOH) documented preadmission and postdischarge were more evident for the disabled population than the aged. Furthermore, when pre- and postadmission spans for SDOH conditions were examined, (6 months before initial admission to 30 days postadmission), both the aged and disabled populations were statistically significantly more likely to experience readmissions if they had an SDOH diagnosis.

Introduction

Hospital readmissions are not only costly but they are also reported as a measure of quality of care provided to patients. Readmissions may be due to a variety of causes including severe disease, incomplete care or treatment, suboptimal health behavior (eg, noncompliance), poor coordination of services at discharge, or unfavorable social determinants of health (SDOH). Published estimates vary, however, a study in 2016 reported that over a quarter of all readmissions were potentially avoidable in the general population.¹ Furthermore, a study examining rehospitalization events found that among Medicare beneficiaries, ∼20% were readmitted within 30 days of index hospitalization, and 34% were rehospitalized within 90 days of discharge.² The Agency for Healthcare Research and Quality (AHRQ) reports that readmissions are often more costly than an index admission event (https://www.hcup-us.ahrq.gov/reports/statbriefs/sb248-Hospital-Readmissions-2010-2016.jsp). In an effort to combat rising readmission rates and improve patient outcomes and health care, the Centers for Medicare and Medicaid Services (CMS) instituted the Hospital Readmissions Reduction Program (HRRP) in FY 2013.³ This program targets the following 6 conditions in an effort to reduce 30-day unplanned readmissions for the Medicare beneficiary population: acute myocardial infarction, chronic obstructive pulmonary disease (COPD), heart failure, pneumonia, coronary artery bypass graft (CABG) surgery, elective primary total hip arthroplasty, and/or total knee arthroplasty.

Understanding the association between personal and historical patient factors and identifying individuals most at risk for readmission are important to improving patient care and decreasing utilization of costly inpatient services. Doing so can provide an opportunity to intervene with the riskiest patients in a preventive manner. Previous studies identified methods to predict hospital stays that are likely to result in readmissions including multiple chronic conditions, premature discharge from hospital, and length of stay of index admission.² The majority of these studies focused on specific disease states such as heart attack, heart failure, pneumonia, COPD, elective hip and knee replacement, and CABG as defined by the Centers for Medicare and Medicaid's HRRP. The studies then commonly concentrated on hospital activities such as early discharge and length of stay as predictive variables. Most studies reviewed employed hospital electronic medical record (EMR) data in their analysis, the few that included claims data focused on specific conditions.

However, current modeling based on EMR data, such as the Length of Stay (L), Acuity of Admission (A), Comorbidity if Patient (C) and Emergency (E) index (which uses hospital records to predict readmissions by examining the length of stay, acuity of illness, comorbid conditions, and emergency department visits), fails to consider other important contributors such as SDOH data and an individual's health history.⁴ Most current models to identify individuals at risk for readmission have generally been found to be poor indicators of readmission risk,⁵ and seem to have the most predictive power when focused on 1 specific disease state, and are thus inapplicable to individuals with a different disease state, or multiple conditions. Because most studies have focused on the index admission facts for a specific disease state only, they often are not generalizable to the overall contributors of readmission. In addition, although an emphasis on Medicare is reasonable due to the impact of readmission on that population, these studies focused only on the aged Medicare population, and did not include those individuals who receive Medicare benefits due to disability status.

Currently, readmission risk prediction is not well understood, often ill defined, and future research needs to take into account broader information than a single disease state.⁶ Indeed, several studies highlighted key gaps to be filled by future studies. These gaps include socioeconomic factors, mental health issues, and drug abuse.⁵ Therefore, using claims data to predict readmissions may be a useful complement to existing approaches that rely on EMR data, as claims data are able to incorporate a comprehensive patient history regardless of provider, whereas hospital EMR data are event focused. Moreover, claims data may provide the opportunity to systematically identify factors (such as historical trends, prior treatment and treatment patterns, and comorbidities) that are not present in the acute care EMR. Claims data offer the opportunity to review a comprehensive history of a Medicare beneficiary's health status by identifying comorbid conditions, his/her resource utilization patterns, disease management and preventive care activities, and documented SDOH. SDOH are defined as the conditions in which one lives, such as social, economic, physical environments, and cultural and behavioral factors, including social support systems that contribute to health status.⁷

Methods

Inpatient admissions were identified from a 5% national sample Medicare Fee For Service claims records from 2014 to 2019. Those with end-stage renal disease (ESRD) were excluded to limit to those who had readmissions that were preventable. Similarly, additional exclusion criteria implemented were select primary International Classification of Diseases (ICD) diagnoses (neoplasm, pregnancy, childbirth, or perinatal), admission type (newborn or unknown), discharge to another care facility for continued care (eg, skilled nursing facility, Medicare swing bed, inpatient rehabilitation facility), patient discharge status (expired or unknown), and patients who died within 30 days of discharge. Most exclusions were considered because they were indicative of a situation that could not be managed to prevent a readmission. Discharge against medical advice (AMA) was retained in the model, contrary to CMS measure for hospital quality, because this model does not reflect on the provider but rather on the patient, for whom AMA is a choice. An algorithm was used to determine an entire stay or “admit episode” from acute inpatient admission claims. Claims from the same member and provider were sorted by date, then, if a claim started on the same date as another's end date, it was determined that they were a part of the same episode. After all claims were examined, length of stay was calculated from the earliest date to the latest date for each admit episode. Primary diagnosis was identified as discharge diagnosis when available or earliest primary diagnosis. The final data set had 1,271,152 inpatient admission episodes for 637,745 unique individuals for 6 years. Thirty-day all-cause readmission was defined as an inpatient admission within 30 days of the latest date of the initial admit episode. This study identified readmissions for any reason and, therefore, is not identical to the CMS 30-day all-cause hospital readmission measure calculated for the value-based payment modifier program, which measures the quality of care by physicians⁸

Initial exploration of the data showed that 2 different populations should be stratified into distinct groups for analysis by original Medicare eligibility reason: aged (459,066 unique individuals and 823,847 admit episodes) or disability (178,679 unique individuals and 447,305 admit episodes). Those Medicare recipients who became eligible due to a disability are under age 65 years and received Social Security Disability benefits for 24 months or have ESRD or amyotropic lateral sclerosis (also known as Lou Gehrig's disease). For the aged population, 11.8% of the acute inpatient admissions resulted in a readmission within 30 days, whereas 17.8% was true of the disabled population. These distinct groups are defined in the Medicare enrollment data. The aged population had a higher rate of elective admissions (23.3%) than the disabled population (16.9%), likely due to the disabled population having a greater incidence and severity of illness due to eligibility requirements. Similarly, the aged population experienced a greater percentage of discharges with a home health service organization than the disabled population (30.2% and 21.7%, respectively). Chi-square tests of independence and Mann–Whitney U tests were performed to assess statistically significant differences in the admissions between the 2 groups. Results of that test are found in Table 1. This study was conducted with the approval of an IRB HSC-SPH-19-0730.

Table 1.

2015–2019 Medicare Admissions

Admissions for all members N = 1,271,152	Aged member admissions		Disabled member admissions		χ²	P
Admissions for all members N = 1,271,152	N	%	N	%	χ²	P
	823,847	—	447,305	—
Gender					2134.3	<0.0001
Male	373,373	45.3	221,873	49.6
Female	450,474	54.7	225,432	50.4
Race					45,496.9	<0.0001
White	714,631	86.7	331,603	74.1
Black	61,539	7.5	85,063	19.0
Asian	11,412	1.4	3474	0.8
Hispanic	12,850	1.6	14,354	3.2
NA native	4607	0.6	5114	1.1
Other	9527	1.2	4271	1.0
Unknown	9281	1.1	3426	0.8
Age					70,0204.0	<0.0001
18–24	0	0.0	4005	0.9
25–34	0	0.0	24,918	5.6
35–44	0	0.0	44,374	9.9
45–54	0	0.0	81,850	18.3
55–64	0	0.0	138,852	31.0
65+	823,847	100.0	153,306	34.3
Admission type					9052.2	<0.0001
Emergency	517,199	62.8	292,936	65.5
Urgent	109,391	13.3	76,590	17.1
Elective	192,064	23.3	75,407	16.9
Trauma center	5193	0.6	2372	0.5
Admission source					3855.5	<0.0001
Referral	761,718	92.5	399,527	89.3
Transfer	59,728	7.2	45,020	10.1
Unknown	2401	0.3	2758	0.6
Discharge status					19,279.5	<0.0001
Home	557,706	67.7	336,756	75.3
Home health service	249,097	30.2	96,930	21.7
Against medical advice	4576	0.6	10,360	2.3
Hospice	12,468	1.5	3259	0.7
SDOH DX at admission					13,863.9	<0.0001
Yes	3580	0.4	13,063	2.9
No	820,267	99.6	434,242	97.1
Behavioral health DX at admission					74,909.7	<0.0001
Yes	62,653	7.6	112,364	25.1
No	761,194	92.4	334,941	74.9
Behavioral health DX before 6 months					185.5	<0.0001
Yes	35,107	4.3	50,927	11.4
No	788,740	95.7	396,378	88.6
SDOH DX 6 months before 30 days postdischarge					11,679.5	<0.0001
Yes	10,794	1.3	19,570	4.4
No	813,053	98.7	427,735	95.6

	Mean	SD	Mean	SD
Mean no. of inpatient visits 6 months prior	0.618	1.012	1.034	1.605	20,663.3	<0.0001
Mean no. of emergency department visits 6 months prior	0.876	1.745	2.184	4.153	62,212.8	<0.0001
Mean length of stay (days)	4.570	5.348	5.692	7.143	13,746.6	<0.0001

DX, diagnosis; NA, Native American; SDOH, social determinants of health.

Fourteen continuous and categorical variables were available for input into the model. Continuous variables were age (years), number of inpatient admissions 6 months before episode, number of emergency visits 6 months before episode, and episode length of stay (days). Categorical variables were gender, race, admission type, admission source, discharge status, social factors determining health diagnosis at admission, social factors determining health diagnosis 30 days before or 6 months before episode, behavioral or drug abuse diagnosis at admission, behavioral or drug abuse diagnosis 6 months before episode, and principal diagnosis. Some examples of SDOH included psychosocial and socioeconomic circumstances related to education, literacy, income, housing, support groups, and family stressors. SDOH variables were identified through ICD-10 diagnosis codes ranging from Z55 to Z65. Behavioral health diagnoses included psychological disturbances and physical factors and disorders due to psychoactive substance abuse. Refer to Appendix Table A1 for a full list of codes. Age was excluded from the aged population models due to its relation to Medicare eligibility.

All analyses and statistic calculations were performed using SAS 9.4 software^© 2016 SAS Institute, Inc., Cary, NC. Because statistical significance can be easily achieved with a large data set, all variables were first entered into a decision tree using the HPSPLIT procedure, which is a high-performance procedure for regression. The model is expressed as a series of if–then statements that the HPSPLIT procedure then uses to determine the importance and impact variables had on the likelihood of 30-day readmission. Variables with a relative importance of 0.05 from the output of the decision tree analysis were then included in a stepwise logistic regression. Significance level 0.05 was required for entry into the model and 0.01 was the significance required to stay in the model. The final model after selection included 12 variables for the aged population and did not include social factor determining health diagnosis at admission and behavior or drug abuse diagnosis at admission. The disabled population cohort model only included 10 variables and did not include gender, race, and admission sources.

Results

Tables 2 and 3 show the differences in inpatient admissions that resulted in a readmission within 30 days by each population group. Roughly 12% of admissions for the aged population resulted in a readmission within 30 days, whereas this was true for ∼18% of the disabled population. Within these readmissions, the aged population had a greater percentage of elective admissions than the disabled population (15.03% and 11.57%, respectively), likely due to the disabled population having greater incidence and severity of illness due to their reason for eligibility. Similarly, the aged population experienced a greater percentage of discharges with a home health service organization than the disabled population (35.9% and 23.45%).

Table 2.

2015–2019 Medicare Admissions for Aged Entitlement Population

Aged population Unique patients = 459,066	Visits that resulted in 30-day readmission			Visits that did not result in 30-day readmission			Total
Aged population Unique patients = 459,066	N	Row %	Column %	N	Row %	Column %	N	%
	97,108	11.8	—	726,739	88.2	—	823,847	—
Gender
Male	44,965	12.0	46.30	328,408	88.0	45.2	373,373	45.3
Female	52,143	11.6	53.70	398,331	88.4	54.8	450,474	54.7
Race
White	82,918	11.6	85.39	631,713	88.4	86.9	714,631	86.7
Black	8641	14.0	8.90	52,898	86.0	7.3	61,539	7.5
Asian	1307	11.5	1.35	10,105	88.5	1.4	11,412	1.4
Hispanic	1733	13.5	1.78	11,117	86.5	1.5	12,850	1.6
NA Native	549	11.9	0.57	4058	88.1	0.6	4607	0.6
Other	1057	11.1	1.09	8470	88.9	1.2	9527	1.2
Unknown	903	9.7	0.93	8378	90.3	1.2	9281	1.1
Age (years)
18–24		0.0	0.00		0.0	0.0	0	0.0
25–34		0.0	0.00		0.0	0.0	0	0.0
35–44		0.0	0.00		0.0	0.0	0	0.0
45–54		0.0	0.00		0.0	0.0	0	0.0
55–64		0.0	0.00		0.0	0.0	0	0.0
65+	97,108	11.8	100.00	726,739	88.2	100.0	823,847	100.0
Admission type
Emergency	68,129	13.2	70.16	449,070	86.8	61.8	517,199	62.8
Urgent	13,892	12.7	14.31	95,499	87.3	13.1	109,391	13.3
Elective	14,593	7.6	15.03	177,471	92.4	24.4	192,064	23.3
Trauma center	494	9.5	0.51	4699	90.5	0.6	5193	0.6
Admission source
Referral	89,604	11.8	92.27	672,114	88.2	92.5	761,718	92.5
Transfer	7200	12.1	7.41	52,528	87.9	7.2	59,728	7.2
Unknown	304	12.7	0.31	2097	87.3	0.3	2401	0.3
Discharge status
Home	60,754	10.9	62.56	496,952	89.1	68.38	557,706	67.7
Home health service	34,859	14.0	35.90	214,238	86.0	29.48	249,097	30.2
Against medical advice	935	20.4	0.96	3641	79.6	0.50	4576	0.6
Hospice	560	4.5	0.58	11,908	95.5	1.64	12,468	1.5
SDOH DX at admission
Yes	598	16.7	0.62	2982	83.3	0.4	3580	0.4
No	96,510	11.8	99.38	723,757	88.2	99.6	820,267	99.6
Behavioral health DX at admission
Yes	8274	13.2	8.52	54,379	86.8	7.5	62,653	7.6
No	88,834	11.7	91.48	672,360	88.3	92.5	761,194	92.4
Behavioral health DX prior 6 months
Yes	4666	13.3	4.80	30,441	86.7	4.2	35,107	4.3
No	92,442	11.7	95.20	696,298	88.3	95.8	788,740	95.7
SDOH DX 6 months before 30 days post discharge
Yes	2042	18.9	2.10	8752	81.1	1.2	10,794	1.3
No	95,066	11.7	97.90	717,987	88.3	98.8	813,053	98.7

	Mean	SD		Mean	SD		Mean	SD
Mean no. inpatient visits 6 months prior	1.076	1.487	—	0.557	0.912	—	0.618	1.012
Mean no. emergency department visits 6 months prior	1.611	3.143	—	0.778	1.431	—	0.876	1.745
Mean length of stay (days)	5.356	5.915	—	4.455	5.259	—	4.570	5.348

DX, diagnosis; NA, Native American; SD, standard deviation; SDOH, social determinants of health.

Table 3.

2015–2019 Medicare Admissions for Disabled Entitlement Population

Disabled population Unique patients = 178,679	Visits that resulted in 30-day readmission			Visits that did not result in 30-day readmission			Total
Disabled population Unique patients = 178,679	N	Row %	Column %	N	Row %	Column %	N	%
	79,453	17.8	—	367,852	82.2	—	447,305	—
Gender
Male	41,212	18.6	51.87	180,661	81.4	49.1	221,873	49.6
Female	38,241	17.0	48.13	187,191	83.0	50.9	225,432	50.4
Race								0.0
White	57,310	17.3	72.13	274,293	82.7	74.6	331,603	74.1
Black	16,596	19.5	20.89	68,467	80.5	18.6	85,063	19.0
Asian	553	15.9	0.70	2921	84.1	0.8	3474	0.8
Hispanic	2747	19.1	3.46	11,607	80.9	3.2	14,354	3.2
NA native	888	17.4	1.12	4226	82.6	1.1	5114	1.1
Other	752	17.6	0.95	3519	82.4	1.0	4271	1.0
Unknown	607	17.7	0.76	2819	82.3	0.8	3426	0.8
Age (years)
18–24	1075	26.8	1.35	2930	73.2	0.8	4005	0.9
25–34	6061	24.3	7.63	18,857	75.7	5.1	24,918	5.6
35–44	9477	21.4	11.93	34,897	78.6	9.5	44,374	9.9
45–54	15,666	19.1	19.72	66,184	80.9	18.0	81,850	18.3
55–64	23,192	16.7	29.19	115,660	83.3	31.4	138,852	31.0
65+	23,982	15.6	30.18	129,324	84.4	35.2	153,306	34.3
Admission type
Emergency	55,718	19.0	70.13	237,218	81.0	64.5	292,936	65.5
Urgent	14,237	18.6	17.92	62,353	81.4	17.0	76,590	17.1
Elective	9195	12.2	11.57	66,212	87.8	18.0	75,407	16.9
Trauma center	303	12.8	0.38	2069	87.2	0.6	2372	0.5
Admission source
Referral	70,238	17.6	88.40	329,289	82.4	89.5	399,527	89.3
Transfer	8678	19.3	10.92	36,342	80.7	9.9	45,020	10.1
Unknown	537	19.5	0.68	2221	80.5	0.6	2758	0.6
Discharge status
Home	57,507	17.1	72.38	279,249	82.9	75.91	336,756	75.3
Home health service	18,630	19.2	23.45	78,300	80.8	21.29	96,930	21.7
Against medical advice	3018	29.1	3.80	7342	70.9	2.00	10,360	2.3
Hospice	298	9.1	0.38	2961	90.9	0.80	3259	0.7
SDOH DX at admission
Yes	3454	26.4	4.35	9609	73.6	2.6	13,063	2.9
No	75,999	17.5	95.65	358,243	82.5	97.4	434,242	97.1
Behavioral health DX at admission
Yes	22,571	20.1	28.41	89,793	79.9	24.4	112,364	25.1
No	56,882	17.0	71.59	278,059	83.0	75.6	334,941	74.9
Behavioral health DX prior 6 months
Yes	8618	16.9	10.85	42,309	83.1	11.5	50,927	11.4
No	70,835	17.9	89.15	325,543	82.1	88.5	396,378	88.6
SDOH DX 6 months before 30 days postdischarge
Yes	5592	28.6	7.04	13,978	71.4	3.8	19,570	4.4
No	73,861	17.3	92.96	353,874	82.7	96.2	427,735	95.6

	Mean	SD		Mean	SD		Mean	SD
Mean no. inpatient visits 6 months prior	1.970	2.401	—	0.832	1.288	—	1.034	1.605
Mean no. emergency department visits 6 months prior	4.168	6.505	—	1.756	3.286	—	2.184	4.153
Mean length of stay (days)	6.319	7.024	—	5.556	7.161	—	5.692	7.143

DX, diagnosis; NA, Native American; SD, standard deviation; SDOH, social determinants of health.

The most notable difference between the two populations was the increased presence of SDOH factors for the individuals in the disabled population at admission (4.35% compared with 0.62% in the aged group) and 6 months before 30 days postdischarge (7.04%–2.10% in the aged group). Additional differences between groups included behavioral or drug abuse diagnosis at admission, and behavioral or drug abuse diagnosis in the 6 months before initial admission in the disabled population (28.41% and 10.85%) than the aged population (8.52% and 4.80%).

Table 4 provides the odds ratio for significant variables for each respective eligible population. Variables that were not significant were omitted from the tables (ie, age, SDOH diagnosis at admission, behavioral health DX at admission for the aged population, gender, race, SDOH DX at admission, admission source for the disabled population). C-statistics were calculated to compare the models including social determinant and behavioral factors (after selection processes) versus models that did not include those factors. To measure goodness-of-fit, expected outcomes are calculated using risk scores from the model results and compared to the true observed outcomes. The percentage of correctly predicted outcomes from the model is the c-statistic which was used to compare models. For the aged population, the model that did not include any SDOH or behavioral factors had a c-statistic 0.665, whereas the model that did include those factors had c-statistic 0.666. Similarly, for the disabled population, the model that did not include SDOH or behavioral factors had a c-statistic of 0.691, whereas the model that did include the factors had c-statistic 0.695. Although those are very small percentage changes, the increase in number of correctly identified observation pairs was very large due to the size of the sample population.

Table 4.

Odds Ratios of Variables Impacting Readmission by Medicare Eligible Populations

Variable	Level	Aged			Disabled
Variable	Level	OR	[95% CI]	Significant	OR	[95% CI]	Significant
No. of emergency visits	—	1.129	[1.12–1.13]	^*	1.325	[1.32–1.33]	^*
No. of inpatient admissions	—	1.253	[1.24–1.26]	^*	1.043	[1.04–1.05]	^*
Primary diagnosis category	Behavioral vs. circulatory	0.972	[0.92–1.02]		1.176	[1.14–1.21]	^*
Primary diagnosis category	Blood or immune vs. circulatory	1.432	[1.36–1.51]	^*	1.109	[1.05–1.18]	^*
Primary diagnosis category	Congenital vs. circulatory	0.844	[0.67–1.07]		0.925	[0.71–1.21]
Primary diagnosis category	Digestive vs. circulatory	0.971	[0.95–0.99]	^*	1.003	[0.97–1.04]
Primary diagnosis category	Endocrine vs. circulatory	1.078	[1.04–1.12]	^*	1.026	[0.99–1.07]
Primary diagnosis category	ENT vs. circulatory	0.480	[0.39–0.59]	^*	0.423	[0.31–0.58]	^*
Primary diagnosis category	Genitourinary vs. circulatory	1.073	[1.04–1.10]	^*	0.953	[0.92–0.99]	^*
Primary diagnosis category	Infectious vs. circulatory	0.984	[0.96–1.01]		0.955	[0.92–0.99]	^*
Primary diagnosis category	Injury/poison/other vs. circulatory	0.787	[0.76–0.81]	^*	0.842	[0.81–0.88]	^*
Primary diagnosis category	MSK vs. circulatory	0.487	[0.47–0.51]	^*	0.552	[0.53–0.58]	^*
Primary diagnosis category	Nervous vs. circulatory	0.899	[0.86–0.94]	^*	0.810	[0.77–0.85]	^*
Primary diagnosis category	Other abnormal vs. circulatory	0.849	[0.82–0.88]	^*	0.804	[0.77–0.84]	^*
Primary diagnosis category	Respiratory vs. circulatory	1.055	[1.03–1.08]	^*	1.021	[0.99–1.05]
Primary diagnosis category	Z code vs. circulatory	1.809	[1.68–1.95]	^*	1.381	[1.23–1.55]	^*
Primary diagnosis category	Skin vs. circulatory	0.840	[0.80–0.89]	^*	0.839	[0.79–0.89]	^*
Length of stay (days)	—	1.017	[1.02–1.02]	^*	1.007	[1.01–1.01]	^*
Discharge status	Home health vs. discharged home	1.242	[1.22–1.26]	^*	1.274	[1.25–1.30]	^*
Discharge status	Against medical advice vs. discharged home	1.678	[1.55–1.81]	^*	1.566	[1.49–1.64]	^*
Discharge status	Hospice vs. discharged home	0.267	[0.25–0.29]	^*	0.433	[0.38–0.49]	^*
Admission type	Emergency vs. elective	1.377	[1.35–1.41]	^*	1.300	[1.27–1.34]	^*
Admission type	Traumacenter vs. elective	1.157	[1.05–1.28]	^*	1.060	[0.93–1.21]
Admission type	Urgent vs. elective	1.379	[1.34–1.42]	^*	1.231	[1.19–1.27]	^*
Race	Asian vs. White	0.930	[0.88–0.99]	^*
Race	Black vs. White	1.071	[1.04–1.10]	^*
Race	Hispanic vs. White	1.066	[1.01–1.12]	^*
Race	Native vs. White	1.036	[0.95–1.14]
Race	Other vs. White	0.936	[0.88–1.00]	^*
Race	Unknown vs. White	0.900	[0.84–0.97]	^*
Admission source	Transfer vs. referral	0.908	[0.88–0.93]	^*
Admission source	Unknown vs. referral	0.994	[0.88–1.13]
Gender	Female vs. male	0.942	[0.93–0.96]	^*
Prior behavioral diagnosis	Yes vs. no	0.817	[0.79–0.85]	^*	0.680	[0.66–0.70]	^*
SDOH diagnosis 6 months before 30 days after	Yes vs. no	1.274	[1.21–1.34]	^*	1.297	[1.25–1.35]	^*
Behavioral diagnosis at admission	Yes vs. no				1.063	[1.04–1.09]	^*
Age at admission (years)	—				0.998	[1.00–1.00]	^*

Significance of ^* is P < 0.05.

Z codes are factors influencing health status and contact with health services.

ENT, Ear, Nose and Throat; MSK, Musculoskeletal; SDOH, social determinants of health.

In review of personal demographics, the researchers found that age did not seem to play a significant role in readmission risk in the disabled Medicare beneficiary group. However, it is of interest that within the disabled beneficiary group, where there is a wide distribution since this group includes ages 18–65 years and over, age was still a less important factor in readmission contribution than other variables of interest. In the aged population, females were less likely to experience a readmission than their male counterparts (OR 0.94, 95% CI 0.93–0.96). In the aged population, race also had a slight predictive effect for readmission risk with Black (OR 1.07, 95% CI 1.04–1.10) and Hispanic (OR 1.066, 95% CI 1.01–1.12) at greater odds than their White counterparts. Asians were less likely to experience readmissions (OR 0.93, 95% CI 0.88–0.99).

Factors related to the admission itself were significant for both populations, confirming findings of other admission-specific studies. For the aged population, an urgent or emergency admission increased the odds of readmission (OR 1.379, 95% CI 1.34–1.42 and OR 1.377, 95% CI 1.35–1.41, respectively) and the same factors increased the odds of readmission for the disabled population (OR 1.231, 95% CI 1.19–1.27 and OR 1.300, 95% CI 1.27–1.34). Also, for both groups, a discharge AMA greatly increased the odds of readmission (OR 1.678, 95% CI 1.55–1.81 for aged and OR 1.566, 95% CI 1.49–1.64 for disabled), as did a discharge to home health services (OR 1.242, 95% CI 1.22–1.26 for aged; OR 1.274, 95% CI 1.25–1.30 for disabled). Interestingly, contrary to what is cited in some of the literature, length of stay during the index admission did not seem to greatly increase a beneficiary's risk of readmission. As might be expected, the primary admit diagnosis contributed to the odds of readmission for both groups. For the aged, a diagnosis of a blood or immune condition increased the odds by 43%, and a behavioral health condition for the disabled increased the odds of 17%. For both groups, a primary admit diagnosis that fell within the category of Z codes, defined as factors influencing health and contact with health services, had the highest odds ratio among diagnoses (OR 1.809, 95% CI 1.68–1.95 for aged; OR 1.381, 95% CI 1.23–1.55 for disabled).

Unique to this study are the findings related to the impact of patient history or resource utilization and noted diagnoses. Frequency of hospital inpatient admissions was discovered to be an important indicator of subsequent readmission risk; the number of inpatient stays within 6 months of the index stay was associated with a significantly higher likelihood of readmission in both beneficiary groups (OR 1.25, 95% CI 1.24–1.26 in the aged and OR 1.33, 95% CI 1.32–1.33 in the disabled). Similarly, the number of emergency department visits in the 6 months before admission increased the odds of readmission for both groups (OR 1.129, 95% CI 1.12–1.33 in the aged and OR 1.043, 95% CI 1.01–1.01 in the disabled). Evidence of a behavioral health diagnosis in the 6 months before admission reduced the likelihood of readmission for both groups (OR 0.817, 95% CI 1.2–1.34 in the aged and OR 0.68, 95% CI 0.66–0.70 in the disabled).

Finally, and perhaps most importantly, we found that a recorded diagnosis of an SDOH within 6 months before the admission, during the admission, or up to 30 days after the admission increased the odds of readmission by 27% for the aged and 29% for the disabled. For the purposes of this examination into readmission risk, SDOH codes refer to a group of diagnosis codes that identify persons with potential health hazards related to socioeconomic and psychosocial circumstances. These codes range from problems related to education and literacy, employment issues, housing circumstances, and occupational exposure to risk factors, among others. The list of codes is provided in Appendix Table A1.

Discussion

Zhou et al⁹ conducted a systematic review (2011–2015) of the literature relating to risk predictive models for unplanned hospital readmissions. They found that overall, a wide range of c-statistics were reported in 56 of 60 studies, and only 11 of 13 predictive models for specific medical condition-related readmissions were found to have consistent moderate discrimination ability (c-statistic ≥0.7). This study produces a c-statistic just <0.7 for all models for each of the populations.

A key finding from this study is that the 2 cohorts of Medicare beneficiaries differ both in eligibility requirements and in the factors that contribute to readmissions. The aged, who are eligible for Medicare when they turn 65 years, had higher odds for readmission based on the clinical condition associated with the index admission. For example, inpatient admissions with primary diagnosis of blood or immune disorders had a 43% greater likelihood of readmission than the reference condition of circulatory disease. For the disabled population, the physical factor that had the greatest impact was behavior health diagnoses, which were associated with a 17.6% increased likelihood of readmission. Several factors associated with the admission itself, such as admission type and discharge status contributed greatly for both cohorts. Furthermore, social factors also play an important role in readmission risk for both populations, as demonstrated by these findings. These findings stress that the risk of readmission exists for many clinical conditions other than those focused on by CMS, and that the entire inpatient experience from source of admission to discharge status as well as personal living conditions has an impact on readmission risk.

These are important findings to consider because, as discovered in a review of the literature, most existing models only examine a specific set of conditions, such as those originally identified by the HRRP, and are limited to treatment events only.³ These previous models failed to take into consideration SDOH and other behavioral issues,⁵ which we found to be significant indicators for readmission risk. SDOH codes are not frequently documented in a claim as they have no influence over payment, so it is especially interesting that they demonstrated significant contributing amounts. A very recent study by Bensken et al¹⁰ also reviewed the importance of social determinants on readmission, using from ICD-10 diagnosis codes. Similar to this study, they found that although SDOH was not frequently documented, it did significantly increase the odds of readmission.

Models to predict readmission risk in the Medicare beneficiary population should focus on individuals with an emergent or urgent index hospital event as well as where the patient was discharged to from the initial inpatient event. In addition, prediction models should direct attention to the existence of SDOH factors for the individual, which greatly contributed to the odds of readmission. This analysis determined that these variables were most strongly associated with a 30-day hospital readmission.

Conclusions

It is important to note that the 2 populations that comprise the Medicare beneficiary group as a whole are quite different. As noted, both the aged and the disabled populations experienced increased readmission risk when SDOH conditions were recorded. These codes are indicative of potential health hazards related to socioeconomic and psychosocial circumstances and may be indicators for living situations or personal factors that may need intervention to decrease readmission risk. Further research may be needed to further quantify the impact of social determinants on readmission risk in Medicare beneficiaries. These results are specific to the Medicare beneficiary population and are likely not generalizable to a younger population such as those covered by Medicaid, or even commercially insured individuals. It is likely that if replicated for either Medicaid or commercially insured populations, the results and factors identified as contributing to readmissions would be very different.

A potential limitation is the use of only claims data in this study. Although there are some advantages to claims data such as the captured history, they may pose a challenge to hospital systems and providers who would like to apply the model. In addition, Donzé et al¹¹ found that administrative data may not always accurately identify elective from nonelective readmissions.

This study generated equations from the logistic regressions for each population that may be applied to assess risk for readmission to future cases upon the initial admission. This might be a valuable tool to identify persons at high risk to whom possible interventions might be directed for prevention. Future studies will evaluate the validity and usefulness of these equations for the 2 populations eligible for Medicare.

Footnotes

Author Disclosure Statement

No competing financial interests exist.

Funding Information

This research was funded by the Cullen Foundation. There is no grant number.

Appendix

Appendix Table A1.

List of Diagnoses Codes

ICD-10	ICD-9	Model category	Official category
A00–B99	001–139	Infectious	Certain infectious and parasitic diseases
C00–D49	140–239	Neoplasm	Neoplasms
D50–D89	280–289	Blood or immune	Diseases of the blood and blood-forming organs and certain disorders involving the immune mechanism
E00–E89	240–279	Endocrine	Endocrine, nutritional, and metabolic diseases
F01–F99	290–319	Mental/behavioral	Mental, behavioral, and neurodevelopmental disorders
G00–G99	320–389	Nervous	Diseases of the nervous system
H00–H59	—	ENT	Diseases of the eye and adnexa
H60–H95	—	ENT	Diseases of the ear and mastoid process
I00–I99	390–459	Circulatory	Diseases of the circulatory system
J00–J99	460–519	Respiratory	Diseases of the respiratory system
K00–K95	520–579	Digestive	Diseases of the digestive system
L00–L99	680–709	Skin	Diseases of the skin and subcutaneous tissue
M00–M99	710–739	MSK	Diseases of the musculoskeletal system and connective tissue
N00–N99	580–629	Genitourinary	Diseases of the genitourinary system
O00–O9A	630–679	Pregnancy/childbirth	Pregnancy, childbirth, and the puerperium
P00–P96	760–779	Preinatal	Certain conditions originating in the perinatal period
Q00–Q99	740–759	congenital	Congenital malformations, deformations, and chromosomal abnormalities
R00–R99	780–799	Other abnormal	Symptoms, signs, and abnormal clinical and laboratory findings, not elsewhere classified
S00–T88	800–999	Injury/poison/other	Injury, poisoning, and certain other consequences of external causes
V00–Y99	E000–E999	External cause	External causes of morbidity
Z00–Z99	V01–V91	Z code	Factors influencing health status and contact with health services

Codes for any indication of “behavioral health” or “SDOH” for additional covariates
F10–F19	290–294, 300–305	Behavioral/drug use	Mental and behavioral disorders due to psychoactive substance use
F50–F59	306–316	Behavioral/drug use	Behavioral syndromes associated with physiological disturbances and physical factors
F60–F69		Behavioral/drug use	Disorders of adult personality and behavior
Z55–Z65		SDOH	Persons with potential health hazards related to socioeconomic and psychosocial circumstances

SDOH, social determinants of health.

References

Auerbach

, Kripalani

, Vasilevskis

, et al. Preventability and causes of readmissions in a national cohort of general medicine patients. JAMA Intern Med, 2016; 176:484–493.

Berenson

, Paulus

, Kalman

. Medicare's readmissions-reduction program—a positive alternative. N Engl J Med, 2012; 366:1364–1366.

Hospital Readmissions Reduction Program (HRRP). 2020. https://www.cms.gov/Medicare/Medicare-Fee-for-Service-Payment/AcuteInpatientPPS/Readmissions-Reduction-Program Accessed March 15, 2020 .

Goldfield

. Strategies to decrease the rate of preventable readmission to hospital. Can Med Assoc J, 2010; 182:538–539.

Kansagara

, Englander

, Salanitro

, et al. Risk prediction models for hospital readmission: a systematic review. JAMA, 2011; 306:1688–1698.

Kroch

, Duan

, Martin

, Bankowitz

. Patient factors predictive of hospital readmissions within 30 days. J Healthc Qual, 2016; 38:106–115.

Social Determinants of Health. 2020. https://www.healthypeople.gov/2020/topics-objectives/topic/social-determinants-of-health Accessed January 7, 2021 .

2015 Measure information about the 30-day all-cause hospital readmission measure, calculated for the value-based payment modifier program. https://www.cms.gov/Medicare/Medicare-Fee-for-Service-, Payment/PhysicianFeedbackProgram/Downloads/2015-ACR-MIF.pdf Accessed January 10, 2021 .

Zhou

, Della

, Roberts

, et al. Utility of models to predict 28-day or 30-day unplanned hospital readmissions: an updated systematic review. BMJ Open, 2016; 6:e011060.

10.

Bensken

, Alberti

, Koroukian

. Health-related social needs and increased readmission rates: findings from the Nationwide Readmissions Database. J Gen Intern Med, 2021; 36:1173–1180.

11.

Donzé

, Aujesky

, Williams

, Schnipper

. Potentially avoidable 30-day hospital readmissions in medical patients: derivation and validation of a prediction model. JAMA Intern Med, 2013; 173:632–638.