Medical and Social Factors Associated With Referral for Elder Abuse Services in a National Health Care System

Abstract

Background

Elder abuse (EA) is common and has devastating health consequences yet is not systematically assessed or documented in most health systems, limiting efforts to target health care-based interventions. Our objective was to examine sociodemographic and medical characteristics associated with documented referrals for EA assessment or services in a national U.S. health care system.

Methods

We conducted a national case–control study in U.S. Veterans Health Administration facilities of primary care (PC)-engaged Veterans age ≥60 years who were evaluated by social work (SW) for EA-related concerns between 2010 and 2018. Cases were matched 1:5 to controls with a PC visit within 60 days of the matched case SW encounter. We examined the association of patient sociodemographic and health factors with receipt of EA services in unadjusted and adjusted models.

Results

Of 5 567 664 Veterans meeting eligibility criteria during the study period, 15 752 (0.3%) received services for EA (cases). Cases were mean age 74, and 54% unmarried. In adjusted logistic regression models (adjusted odds ratio; 95% confidence interval), age ≥ 85 (3.56 vs age 60–64; 3.24–3.91), female sex (1.96; 1.76–2.21), child as next-of-kin (1.70 vs spouse; 1.57–1.85), lower neighborhood socioeconomic status (1.18 per higher quartile; 1.15–1.21), dementia diagnosis (3.01; 2.77–3.28), and receiving a VA pension (1.34; 1.23–1.46) were associated with receiving EA services.

Conclusion

In the largest cohort of patients receiving EA-related health care services studied to date, this study identified novel factors associated with clinical suspicion of EA that can be used to inform improvements in health care-based EA surveillance and detection.

Elder abuse (EA), including physical, sexual, and psychological abuse, financial exploitation, and neglect, is a common and devastating problem in the United States. Approximately 10% of the 70 million U.S. older adults experience EA annually (1–3), with associated increases in depression (4), emergency department visits (5), hospitalization (6), nursing home placement (7), and mortality (8–10). Unfortunately, most cases go undetected (3), which limits opportunities for interventions that can improve outcomes. Health care visits provide a unique opportunity for EA detection and connection to needed services, particularly for older adults who may be socially isolated (6,11). However, while efforts have increased to detect and address EA in health care settings (12), care that addresses EA concerns in electronic health records (EHR) is not systematically documented (eg, via diagnostic codes) (13). Lack of routine documentation of EA suspicion or assessment in the health care system limits the ability to study and improve health care for older patients at risk for EA.

The ability to study a large population of older adults receiving EA services during routine health care could achieve 2 goals: better characterizing known risk factors for experiencing EA and identifying factors associated with opportunity and likelihood of referring patients for EA services (Figure 1). Conceptual frameworks of risk for experiencing EA delineate 3 types of factors: those related to the older adult, those related to the trusted other, and those related to the context of the relationship between the two (14,15). Prior studies of these risk factors have almost exclusively been collected from surveys done as part of stand-alone research studies and have been limited by small sample sizes (5–7,16), exclusion of those with dementia (1,17,18), and reporting (19) and participation (1,17,20) bias. There are no known studies that systematically describe the older adult population whose EA is detected or addressed during routine clinical care. System-wide data collected in natural health care settings would provide unique information on risk factors for experiencing EA and factors affecting health care-based EA detection and intervention strategies.

Figure 1.

Factors potentially associated with health care-based referral to social work for elder abuse assessment. ED = emergency department.

In 2010, the Veterans Health Administration (VHA), the largest integrated U.S. health system, began systematically collecting data that identifies patients who receive VHA social work (SW) assessment for suspected abuse or neglect, presenting a novel opportunity to examine health care system data related to EA. The VHA also collects robust medical and social information on the approximately 8 million Veterans enrolled for health services, 48% of whom were age ≥65 years in 2017 (21), providing an opportunity to study a wide range of factors associated with clinical suspicion of EA in a large, high-risk population (22). The objective of this study was to assemble a large, national cohort of older adults identified during routine medical care as requiring SW assessment for suspected EA, and to examine patient characteristics associated with receiving EA services that can inform future efforts to improve health care-based EA detection and intervention.

Method

All study activities were approved by the VA Pittsburgh Healthcare System Institutional Review Board.

Sample Population

We drew our sample from Veteran enrollees from 130 VHA facilities who were aged ≥60 years and actively engaged in primary care (PC) between 2010 and 2018.

Case Selection

We identified a cohort of VHA patients completing social work consultations for a referral reason of suspected EA. In the VHA system, patients are typically referred to SW by a PC or other clinical provider who completes an electronic form with free-text entry for the reason for referral. In 2010, SW note templates were implemented nationally that prompt the social worker to mark one of a list of possible referral reasons when a Veteran completes a SW encounter. The social worker completing the consultation selects one or more indications based on their interpretation of the referring provider’s reason for referral. The checked reasons are retrievable as data elements from the VHA national Corporate Data Warehouse (CDW). One referral reason on this list is “abuse/neglect” (referred to henceforth as “EA indicator” when used for patients age 60 and over). The EA indicator does not indicate the type of abuse or neglect that is suspected. There is also not a mechanism to systematically record the outcome of any non-VA investigation into whether the abuse concern is substantiated.

We defined cases as patients who had a completed SW encounter coded with the EA indicator and used the SW encounter date as the index date. We further restricted the sample to patients who had a PC visit within 60 days prior to their index date to capture patients who were actively engaged in PC at the time of the SW encounter to ensure they would have had the opportunity for documentation of key characteristics of interest in the EHR (Figure 2).

Figure 2.

Case and control selection flowchart. VHA = Veterans Health Administration; PC = primary care; SW = social work.

To determine how many cases were documented via more traditional diagnostic markers, we examined the presence of International Classification of Disease (ICD)-9/10 diagnostic codes corresponding to any type of EA (Supplementary Table 1) in visit encounters during the 2 years prior to and including the index date for both cases and controls.

Control Selection

The control group was comprised of patients who did not have the EA indicator or any of 72 other administrative data markers generated at individual VA medical centers (ie, not utilized nationally) identified as potentially related to experience of EA (Supplementary Table 1) during the study period. Five controls for each case were randomly selected from this pool, restricting to those having a PC visit within ±30 days of the case index date. We subsequently excluded controls originating from 15 VHA sites where no EA cases were identified, to avoid misclassification due to the possibility that the templated SW note generating the EA indicator was not actually in use at these facilities (Figure 2).

Exposures

Exposures of interest were selected from conceptual models of EA risk (14,15) based on prior EA risk-factor research (18,19,23), predominantly in the domain of older adult characteristics given the lack of availability of caregiver and contextual level information in EHR data. We did, however, attempt to assess interpersonal characteristics to the extent possible by collecting variables on marital status and documented next-of-kin. In addition to patient characteristics previously demonstrated or hypothesized to be associated with increased risk of experiencing EA, we also included frequency of certain types of health care contacts that may have increased the opportunity for health care providers to detect signs of EA (Figure 1).

Sociodemographic characteristics

We extracted age, sex, race/ethnicity, marital status, rurality of home residence (based on U.S. Census Bureau definitions) and relationship of patient-reported “next-of-kin” from CDW at index, all categorized as shown in Table 1. To measure patients’ neighborhood socioeconomic status (SES) we used the most recent (2015) linked Area Deprivation Index (ADI) from the Neighborhood Atlas (24–26). Prior to modeling, national ADI percentiles were centered about the mean and scaled by the standard deviation.

Table 1.

Characteristics of a National Cohort of Older Veterans With at least One Documented Encounter With Social Work for Abuse/Neglect and Matched Controls

Characteristic, no. (%)	Overall (N = 88 462)	Cases (N = 15 752)	Controls (N = 72 710)
Age category
60–64	20 160 (22.8)	2 870 (18.2)	17 290 (23.8)
65–69	22 499 (25.4)	3 638 (23.1)	18 861 (25.9)
70–74	15 261 (17.3)	2 753 (17.5)	12 508 (17.2)
75–59	9 581 (10.8)	1 737 (11.0)	7 844 (10.8)
80–84	9 580 (10.8)	2 079 (13.2)	7 501 (10.3)
≥85	11 381 (12.9)	2 675 (17.0)	8 706 (12.0)
Male sex	85 315 (96.4)	14 941 (94.9)	70 374 (96.8)
Race/ethnicity
Non-Hispanic White	62 292 (70.4)	9 448 (60.0)	52 844 (72.7)
Non-Hispanic Black	11 889 (13.4)	2 577 (16.4)	9 312 (12.8)
Hispanic	5 063 (5.7)	2 115 (13.4)	2 948 (4.1)
Multiple or another race	2 619 (3.0)	411 (2.6)	2 208 (3.0)
Missing	6 599 (7.5)	1 201 (7.6)	5 398 (7.4)
Marital status
Married	51 204 (57.9)	7 224 (45.9)	43 980 (60.5)
Not marrieda	36 656 (41.4)	8 472 (53.8)	28 184 (38.8)
Missing	602 (0.7)	56 (0.4)	546 (0.8)
Rurality
Urban	54 182 (61.3)	10 069 (63.9)	44 113 (60.7)
Rural/highly rural	32 342 (36.6)	5 400 (34.3)	26 942 (37.1)
Missing	1 938 (2.2)	283 (1.8)	1 655 (2.3)
National area deprivation index mean (SD)	54.7 (25.4)	62.3 (24.5)	53.1 (25.3)
Relationship of next-of-kin
Spouse	44 957 (50.8)	5 639 (35.8)	39 318 (54.1)
Child	20 721 (23.4)	4 965 (31.5)	15 756 (21.7)
Other family	12 843 (14.5)	2 644 (16.8)	10 199 (14.0)
Friend/other	4 236 (4.8)	1 078 (6.8)	3 158 (4.3)
No next-of-kin listed	5 705 (6.5)	1 426 (9.1)	4 279 (5.9)
Reports history of sexual trauma while in military (MST)	2 089 (2.4)	548 (3.5)	1541 (2.1)
Receiving a VA pension	6 939 (7.8)	2 262 (14.4)	4 677 (6.4)
Receiving Aid and Attendance	2 722 (3.1)	1 075 (6.8)	1 647 (2.3)
Receiving housebound benefits	913 (1.0)	309 (2.0)	604 (0.8)
Elixhauser index categories
≤ −5	7 790 (8.8)	1 911 (12.1)	5 879 (8.1)
−4 to −1	7 721 (8.7)	1 859 (11.8)	5 862 (8.1)
0	57 088 (64.5)	6 213 (39.4)	50 875 (70.0)
1–7	8 209 (9.3)	2 331 (14.8)	5 878 (8.1)
≥8	7 654 (8.7)	3 438 (21.8)	4 216 (5.8)
Medical diagnoses
Hearing loss	17 588 (19.9)	3 211 (20.4)	14 377 (19.8)
Vision loss	1 367 (1.6)	540 (3.4)	827 (1.1)
Alcohol abuse	2 785 (3.2)	991 (6.3)	1 794 (2.5)
Drug abuse	1 699 (1.9)	587 (3.7)	1 112 (1.5)
Psychosis	4 579 (5.2)	1 547 (9.8)	3 032 (4.2)
Depression	7 247 (8.2)	2 621 (16.6)	4 626 (6.4)
PTSD	10 782 (12.2)	3 021 (19.2)	7 761 (10.7)
Dementia	4 855 (5.5)	2 335 (14.8)	2 520 (3.5)
Falls	4 325 (4.9)	1 921 (12.2)	2 404 (3.3)
Primary care visits
1–3 visits	31 441 (35.5)	2 440(15.49%)	29 001 (39.9)
4–7 visits	30 473 (34.5)	4 337(27.53%)	26 136 (36.0)
≥8 visits	26 548 (30.0)	8 975 (56.98%)	17 573 (24.2)
Emergency department visits
None	62 050 (70.1)	7 845 (49.8)	54 205 (74.6)
1 visit	10 168 (11.5)	2 081 (13.2)	8 087 (11.1)
2–3 visits	8 566 (9.7)	2 439 (15.5)	6 127 (8.4)
≥4 visits	7 678 (8.7)	3 387 (21.5)	4 291 (5.9)
Mental health visits
None	64 581 (73.0)	8 440 (53.6)	56 141 (77.2)
1–2 visits	7 302 (8.3)	2 160 (13.7)	5 142 (7.1)
3–11 visits	10 235 (11.6)	3 156 (20.0)	7 079 (9.7)
≥12 visits	6 344 (7.2)	1 996 (12.7)	4 348 (6.0)

Notes: MST = military sexual trauma; PTSD = post-traumatic stress disorder. p-Value for all comparisons between cases and controls <.0001 except hearing loss (p = .08).

^aNot married category includes divorced, separated, widow/widower, and never married.

Medical diagnoses

Diagnoses of alcohol abuse, drug abuse, psychosis, and depression were identified by ICD-9 and ICD-10 diagnostic codes assigned to at least 1 inpatient or 2 outpatient encounters (separated by at least 30 days) in VA medical records during the 2 years prior to the index date using modified Elixhauser groupings (27,28). Diagnoses of post-traumatic stress disorder (PTSD) (29), hearing loss (30), and vision loss (31) were obtained based on previously published diagnosis code algorithms. A diagnosis of dementia was determined via diagnostic codes and 3-year look back period as previously validated (32), with minor modifications based on the VHA 2018 Geriatrics and Extended Care Dementia Initiative list of ICD-9/10 codes (Supplementary Table 2). To measure overall comorbidity burden, we used the Elixhauser comorbidity score (33), a weighted summary score of 29 comorbidity conditions validated to predict in-hospital mortality.

The presence of any fall in the 2 years prior to index was examined to represent possible frailty (34–37), and as a potential marker of earlier presentation for injury possibly related to EA. We used published ICD-9 codes for injurious falls (38) and analogous ICD-10 codes identified via an ICD-9/10 crosswalk tool with clinical review for appropriateness (Supplementary Table 3).

VHA health system characteristics

All Veterans enrolled in VHA undergo screening for military sexual trauma (MST), defined as sexual assault and/or repeated, threatening sexual harassment during military service (39). We included MST screening results as previous literature has shown that exposure to prior trauma may increase risk of EA (1).

Certain wartime Veterans with low income and limited net-worth are eligible to receive monthly supplemental pension income from the VA (40). For those with a Veteran pension, additional financial benefits are available: (i) Aid and Attendance, which provides increased monthly pension payments for those requiring help from others for activities of daily living (ADL), and (ii) Housebound benefit, which provides increased monthly pension payments for those who are homebound. We included receipt of these benefits at any time during the 2 years prior to the index date, hypothesizing that the presence of regular income assistance may put older adults at higher risk of financial exploitation and that requiring assistance with ADLs or being homebound may increase the risk of physical abuse and neglect.

Health service utilization

We included health care visit types that we hypothesized were the most likely opportunities for identifying suspicion of EA: PC, mental health, and emergency department (ED). PC visits included in-person or telephone encounters with any provider in the PC setting (eg, physician, nurse, clinical pharmacist). Mental health visits included individual and group visits, in-person or by telephone. We summed visits across the 2 years prior to and including the index date and categorized them based on observed distributions (Table 1).

Statistical Analysis

Sociodemographic characteristics, medical diagnoses, VHA health system characteristics, and health service utilization were described for the whole cohort and by case–control status. We used chi-square and t-tests for bivariate comparisons and plotted the number of cases per year to demonstrate temporal trends in SW visits for EA.

We used logistic regression to examine the unadjusted associations between each patient characteristic listed in Table 1 and receiving SW services for EA. To explore the independent association of each variable with the outcome, we then used one multivariable logistic regression model to examine adjusted associations with receipt of SW services for EA as the dependent variable and all previously examined predictors without variable selection. Given the high number of independent variables, we assessed the data for multicollinearity and removed positive MST screen due to high multicollinearity. Missing data for each variable ranged from 0.4 to 7.6% (Table 1). Missing values for rurality, marital status, ADI, and race/ethnicity were imputed using multiple imputation with chained equations with 10 imputed datasets.(41) All logistic regression models included a fixed effect for VA facility to account for facility-level practice pattern variability. Given the matched case–control study design, conditional logistic regression models were also run as a sensitivity analysis, which yielded similar results. Therefore, we chose to present the output from the standard logistic regression model due to easier interpretation. Two-sided significance was set at p < .05.

We conducted 2 additional sensitivity analyses. First, we conducted additional analyses restricted to cases and controls from 2015 to 2018 due to the steady increase in use of the EA indicator from its inception in 2010 until leveling out in 2015. Second, we conducted sensitivity analyses restricted to cases with ≥2 SW encounters for EA at any time in the study period and their matched controls, as those who had repeat visits may have had elevated clinical suspicion or signs of EA. All analyses were conducted using SAS Enterprise Guide 7.15 and Stata 15.1 (StataCorp LLC, College Station, TX).

Results

Of 5 567 664 Veterans age ≥60 years engaged in VHA PC from 2010 to 2018, 15 752 unique patients (0.3%) from 115 of 130 VHA facilities (88.5%) had at least one SW encounter for abuse/neglect within 60 days following a PC visit (Figure 2). Of Veterans who had at least one SW encounter for EA, there were 3 281 (20.4%) who had 2 or more encounters for a total of 21 205 distinct SW visits for EA during the study period. The incidence of cases identifiable via the EA indicator rose from initial introduction in 2010 until 2015, when the number of new documented EA SW encounters per year plateaued (Supplementary Figure 1).

Table 1 summarizes characteristics of cases with 1 or more EA referral and matched controls. Due to the large sample size, nearly all bivariate comparisons were significant at p < .0001. Overall, cases were older (age ≥80 years 30.2% vs 22.3%), more likely to be female (5% vs 3%), more racially and ethnically diverse (non-Hispanic Black 16.4% vs 12.8%; Hispanic 13.4% vs 4.1%), more likely to be unmarried (53.8% vs 38.8%), and less likely to have a spouse as next-of-kin (35.8% vs 54.1%). Cases had higher mean ADI score (corresponding with lower neighborhood SES) and were more likely to receive a VA pension (14.4% vs 6.4%). Cases had higher prevalence of most medical diagnoses assessed, including depression (16.6% vs 6.4%), PTSD (19.2% vs 10.7%), dementia (14.8% vs 3.5%), and falls (12.2% vs 3.3%). Cases had higher use of all 3 types of health care services assessed compared with controls. Only 0.4% of cases received an EA ICD-9/10 diagnostic code before or on the day of their SW encounter.

Unadjusted odds ratios from bivariable logistic regression models are presented in Table 2. The patterns of association between the unadjusted and adjusted models were largely similar. In the multivariable model, several variables were independently associated with receiving documented SW services for EA (Figure 3). Those age ≥85 years had over 3 times the adjusted odds of EA services compared with those age 60–64 (adjusted odds ratio [aOR] 3.6, 95% confidence interval [CI] 3.2–3.9). Being unmarried (aOR 1.3, 95% CI 1.2–1.4) and having a next-of-kin other than a spouse, or having no next-of-kin, were associated with higher odds of receiving EA services. With each standard deviation (25 points) increase in ADI, odds of receiving EA services were nearly 20% higher (aOR 1.2, 95% CI 1.2–1.2). Cases also had higher odds of receiving all VA financial benefits examined, including VA pension (aOR 1.3, 95% CI 1.2–1.5), Aid and Attendance (aOR 1.5, 95% CI 1.3–1.7), and Housebound benefits (aOR 1.6, 95% CI 1.3–1.9).

Table 2.

Unadjusted and Adjusted Odds Ratios of Characteristics Associated With Receiving Social Work Services for EA (N = 88 462)

Variable	Unadjusted OR (95% CI)	Adjusted OR (95% CI)	p-Value for Adjusted OR
Age group
60–64	Ref	Ref	—
65–69	1.16 (1.10–1.23)	1.54 (1.43–1.67)	<.0001
70–74	1.33 (1.25–1.40)	1.98 (1.82–2.16)	<.0001
75–79	1.33 (1.25–1.42)	2.38 (2.16–2.63)	<.0001
80–84	1.67 (1.57–1.78)	2.95 (2.67–3.24)	<.0001
≥85	1.85 (1.75–1.96)	3.56 (3.24–3.91)	<.0001
Female sex	1.64 (1.51–1.77)	1.97 (1.76–2.21)	<.0001
Race/ethnicity
Non-Hispanic White	Ref	Ref	—
Non-Hispanic Black	1.59 (1.51–1.66)	1.13 (1.04–1.22)	.002
Hispanic	3.93 (3.70–4.17)	1.18 (1.03–1.34)	.015
Other	1.15 (1.03–1.29)	1.20 (1.03–1.40)	.029
Marital status
Married	Ref	Ref	—
Not marrieda	1.82 (1.76–1.89)	1.27 (1.18–1.37)	<.0001
Next of kin
Spouse	Ref	Ref	—
Child	2.20 (2.11–2.29)	1.7 (1.57–1.85)	<.0001
Other family	1.81 (1.72–1.90)	1.64 (1.50–1.81)	<.0001
Friend/other	2.38 (2.21–2.56)	2.25 (2.00–2.53)	<.0001
No next-of-kin	2.32 (2.17–2.48)	1.92 (1.73–2.13)	<.0001
Rurality
Urban	Ref	Ref	—
Rural/highly rural	0.88 (0.85–0.91)	1.09 (1.03–1.15)	.005
ADI (scaled)	1.45 (1.42–1.47)	1.18 (1.15–1.21)	<.0001
Receiving VA pension	2.44 (2.31–2.57)	1.34 (1.23–1.46)	<.0001
Receiving Aid and Attendance	3.16 (2.92–3.42)	1.51 (1.34–1.72)	<.0001
Receiving housebound benefits	2.39 (2.08–2.74)	1.58 (1.29–1.93)	<.0001
Elixhauser index categories
≤ −5	2.66 (2.51–2.82)	1.03 (0.93–1.14)	.60
−4 to −1	2.60 (2.45–2.75)	1.38 (1.27–1.51)	<.0001
0	Ref	Ref	—
1–7	3.25 (3.07–3.43)	1.54 (1.42–1.67)	<.0001
≥8	6.68 (6.34–7.04)	1.86 (1.71–2.02)	<.0001
Dementia	4.85 (4.57–5.14)	3.01 (2.77–3.28)	<.0001
PTSD	1.99 (1.90–2.08)	0.95 (0.88–1.03)	.24
Vision loss	3.09 (2.76–3.44)	1.64 (1.40–1.92)	<.0001
Hearing loss	1.04 (1.00–1.08)	0.86 (0.81–0.92)	<.0001
Falls	4.06 (3.80–4.30)	1.63 (1.49–1.78)	<.0001
Depression	2.94 (2.79–3.09)	1.25 (1.14–1.37)	<.0001
Drug use	2.49 (2.25–2.76)	1.21 (1.04–1.41)	.017
Psychosis	2.50 (2.35–2.67)	1.28 (1.15–1.42)	<.0001
Alcohol use	2.65 (2.45–2.87)	1.10 (0.97–1.24)	.13
Primary care visits
1–3 visits	Ref	Ref	—
4–7 visits	1.97 (1.87–2.08)	1.32 (1.23–1.41)	<.0001
≥8 visits	6.07 (5.78–6.37)	1.94 (1.80–2.09)	<.0001
Emergency department visits
None	Ref	Ref	—
1 visit	1.78 (1.69–1.88)	1.88 (1.75–2.03)	<.0001
2–3 visits	2.75 (2.61–2.90)	2.15 (1.98–2.32)	<.0002
≥4 visits	5.45 (5.18–5.74)	2.99 (2.75–3.26)	<.0001
Mental health visits
None	Ref	Ref	—
1–2 visits	2.79 (2.64–2.95)	2.16 (1.99–2.34)	<.0001
3–11 visits	2.97 (2.83–3.11)	1.93 (1.78–2.10)	<.0001
≥12 visits	3.05 (2.88–3.24)	1.75 (1.56–1.95)	<.0001

Notes: CI = confidence interval; OR = odds ratio.

^aNot married category includes divorced, separated, widow/widower, and never married.

Figure 3.

Forest plot of adjusted odds ratios for characteristics associated with receiving VA social work services for elder abuse. N-H = non-Hispanic; ADI = area deprivation index; PTSD = post-traumatic stress disorder; PC = primary care; ED = emergency department; MH = mental health.

The medical diagnoses most strongly associated with receiving SW services for EA included dementia (aOR 3.0, 95% CI 2.8–3.3), vision loss (aOR 1.6, 95% CI 1.4–1.9), falls (aOR 1.6, 95% CI 1.5–1.8), psychosis (aOR 1.3, 95% CI 1.2–1.4), and depression (aOR 1.3, 95% CI 1.1–1.4). Having diagnoses of PTSD, alcohol use or drug use were not significantly associated with receiving care for EA. Cases had higher adjusted odds of more frequent health service utilization for PC (aOR 1.9 for ≥8 visits per year vs 1–3; 95% CI 1.8–2.1), ED (aOR 3.0 for ≥4 visits per year vs none, 95% CI 2.8–3.3), and mental health visits (aOR 2.2 for 1–2 visits vs none, 95% CI 2.0–2.3).

In the sensitivity analysis restricted to cases and controls from 2015 to 2018 (N = 12 029 cases and N = 54 073 controls), the pattern of associations was similar (Supplementary Table 4). In the sensitivity analysis restricted to 3 210 cases with ≥2 SW visits for EA and 12 264 matched controls, the associations between receiving SW services for EA and race/ethnicity, rurality, drug use, and psychosis became statistically nonsignificant, while the magnitude of some associations became substantively stronger (eg, friend/other next-of-kin aOR 3.5; 95% CI 2.6–4.8; dementia aOR 4.5; 95% CI 3.6–5.7; Supplementary Table 5).

Discussion

In this case–control study with the largest population of older adults identified as receiving health care-based services for EA to date, we identified novel characteristics associated with referral for health care services for EA (eg, low neighborhood SES, receiving a pension) and obtained extensive data on factors associated with experiencing EA that have previously been difficult to study on a large-scale (eg, dementia). Being able to identify patients most likely to need EA services can inform efforts to increase the effectiveness and efficiency of health care system detection, thus increasing the likelihood that those at greatest risk receive EA interventions to improve health outcomes.

With our ability to link comprehensive and systematic EHR data with documented receipt of EA services, we found patterns among factors related to EA care that have not been previously described in past EA research. While prior studies have found an inconsistent relationship of age with EA risk (23,42), we observed higher odds of receiving SW services for EA with each successive age category. One explanation for this finding is that older Veterans may be perceived as more vulnerable and may therefore be more likely to be referred for SW assessment when signs of EA arise. Additionally, the types of EA experienced at older ages (eg, neglect) may be more commonly detected in the health care setting. While some prior studies have found minority racial and ethnic groups may be more likely to experience EA (20,42), others have not found this association (18,43). In primary analyses, we found slightly higher odds of receiving SW services for EA among Black, Hispanic and other non-White identifying races. However, similar to racial biases described in child abuse assessments (44,45), this finding may be the result of biased provider perceptions influencing who is referred for suspected EA rather than true differences in underlying prevalence. Furthermore, the race/ethnicity associations were not statistically significant in the sensitivity analysis using the more restrictive case definition requiring ≥2 SW encounters for EA. This study is the first to examine the association of neighborhood SES with EA services receipt and found that older adults living in neighborhoods with higher deprivation scores were more likely to receive SW services for suspected EA. While previous studies have shown individuals with low income may be at higher risk of experiencing EA (23), these patients may also be more likely to receive care in locations where EA is more actively detected (eg, ED) or to be referred to SW to address other unmet social needs. Our study suggests that those with low resources who qualify for and receive a VA pension may be at particularly high risk for EA, potentially via financial exploitation. We also found that being married and having a spouse as next-of-kin were associated with lower odds of receiving SW services for EA. Conversely, having a nonfamily member next-of-kin was associated with the highest odds, suggesting particular risk among older adults without close family relationships.

Data collected via “secondary” EHR sources may also allow more comprehensive examination of known risk factors for experiencing EA. For example, the extent that dementia represents a risk factor for EA has been difficult to study as most dementia patients are excluded from large EA survey-based prevalence studies (46). In our study, a known dementia diagnosis, as documented via ICD codes in the health care record, was associated with 3 times higher odds of receiving health care services for EA. In the health care setting, individuals with dementia may be less likely to reveal signs of EA due to reduced ability to recognize more subtle forms of abuse (eg, financial) and reduced ability to communicate clearly about EA to providers. Furthermore, patients with dementia are often accompanied by caregivers during health care visits, making direct questioning regarding potential EA challenging. EHR-based EA services data can allow further, in-depth studies of dementia and other EA risk factors that have been previously hard to quantify, while also identifying populations of patients with known risk factors who are not being referred as frequently for EA assessments and may benefit from targeted health system efforts to improve EA-related service delivery. Future studies using EHR data to examine patients who are referred for EA assessments but who do not complete the SW consultation may also help expand our understanding of factors associated with help-seeking behaviors in older adults potentially experiencing EA (47,48).

Our findings represent only those older adults who received EA care that we could identify via a specific administrative data marker. In addition to older Veterans experiencing EA who are not identified, there are likely patients whose EA is detected, and receive EA-related services, but whose care is not documented with the EHR marker used in this study. Indeed, we saw a steady increase in the use of the EA indicator from its initial introduction in 2010, with a steep jump in use in 2014 that corresponded with widespread efforts at the national level within VHA to standardize SW documentation. The existence of this data marker allowed us to identify nearly 16 000 older Veterans referred to SW with suspicion of EA over an 8-year period. Similar to prior studies (13), we found that very few cases received ICD diagnostic codes for EA (0.4%), despite the fact that ICD-10 includes codes for “suspected” abuse and neglect (eTable 1). It is possible that health care providers may recognize abuse and even act on their EA concerns but prefer not to code this diagnosis without an in-depth investigation or avoid documenting given wariness of the legal or patient-safety ramifications of doing so. Additionally, providers may not be aware of the existence of EA-related codes or choose to code immediate medical issues, such as unintentional weight loss or fractures, rather than the underlying cause, which may be neglect or physical abuse. Had we relied on conventional ICD-9/10 diagnostic codes, we would have identified only a small fraction of these patients with clinical suspicion of EA. This shows that diagnostic codes for EA care remain grossly under-used, similar to other social determinants of health (49). Standardizing routine documentation for EA, taking safety and privacy considerations into account, could enable health care systems to better address the health-related social needs of their older patients and assist with EA surveillance. Future studies examining provider or clinic characteristics associated with more identification and documentation of suspected EA will lend important information on these possible determinants of referral for EA services.

This study has several limitations. First, EA is a term that encompasses heterogeneous phenomena; we were not able to differentiate the types of EA suspected in our cases. Different types of EA are likely associated with unique patient characteristics. We were also not able to determine whether clinical suspicions of EA were substantiated. Second, because our EA measure was dependent on provider recognition and referral for EA services, both cases and controls may have been misclassified, and it is unknown to what extent our results reflect provider biases. Third, receiving SW services for EA is dependent on local availability of SW providers and staff to appropriately refer patients, which may vary by site within VHA. We attempted to control for this variability by including facility as a fixed effect in our regression models. Additionally, we were not able to measure other health care-based EA-related services due to the absence of these markers in administrative data. Developing structured data to measure and examine health care services received downstream from SW consultation for EA assessment would be valuable. Fourth, this study included predominantly male, U.S. Veterans; findings may not generalize to non-Veterans or those in other countries. Fifth, it is possible that some of the exposures found to be associated with receiving SW services for EA may in fact result from experience of EA (eg, higher comorbidity and health service utilization), and it was not possible in this study to determine when any potential EA experience started. The characteristics identified in this case–control study cannot be interpreted as causally associated with receiving EA services.

This study demonstrates that integrated health systems have the opportunity to gather data on EA-related services in a systematized way. These data can be used to target EA detection strategies toward populations of older adults with high prevalence of risk factors documented in health care records, such as those with associated diagnoses or types of utilization, and target more comprehensive assessments towards those who risk lack of detection in health care settings, such as those with dementia. These data could also be used to tailor interventions to specific risk-factor profiles, such as financial abuse of those receiving financial benefits. These efforts could then help health systems use existing data to develop practices to proactively detect and intervene upon EA, to improve health outcomes, and to prevent harm among their older patients.

Funding

This work was supported by the U.S. Department of Veterans Affairs, a VA Pittsburgh Center for Health Equity Research and Promotion Competitive Pilot Award (grant number XVA 72-940), and the National Institute on Aging at the National Institutes of Health (grants numbers P30 AG024827 to L.K.M. and K76 AG054866 to T.R.). The content is solely the responsibility of the authors and does not necessarily represent the official views of the U.S. Department of Veterans Affairs or the National Institutes of Health.

Conflict of Interest

None declared.

Author Contributions

Conceptualization: L.K.M., C.T.T., T.R., M.E.D., A.-M.R.; data collection: M.K.M., H.Z., E.L.; data analysis: L.K.M., M.K.M., H.Z., E.L.; data interpretation: L.K.M., C.T.T., M.K.M., T.R., M.E.D., A.-M.R.; writing—original draft preparation: L.K.M.; writing—review and editing: L.K.M., C.T.T., M.K.M., H.Z., E.L., T.R., M.E.D., A.-M.R.; final approval: L.K.M., C.T.T., M.K.M., H.Z., E.L., T.R., M.E.D., A.-M.R. The corresponding author (L.K.M.) attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted.