A Vulnerability Risk Index of Self-Neglect in a Community-Dwelling Older Population

Abstract

BACKGROUND:

The only way to systematically screen for self-neglect among older adults is through in-home observations, which are often difficult and unfeasible for healthcare providers. To fill this gap, we need a robust and efficient prognostication tool to better treat and prevent self-neglect among older adults.

OBJECTIVES:

To develop a predictive index that can be used to assess risk prognostication of the onset of self-neglect among community-dwelling older populations.

DESIGN:

Two waves of longitudinal data from the Chicago Health and Aging Project (CHAP), collected during 2008 to 2012 with approximately 3-year follow-up intervals.

SETTING:

Non-Hispanic black or non-Hispanic white community-dwelling older adults in three adjacent neighborhoods in Chicago, IL.

PARTICIPANTS:

A total of 2885 individuals who were participants of the CHAP study.

MEASUREMENTS:

The main outcomes are incident self-neglect cases. A total of 86 potential predictors were considered in the domains of sociodemographic and socioeconomic, general well-being, health behavior, medical health, medicine/healthcare, cognitive function, physical well-being, social well-being, and psychological well-being.

RESULTS:

The 3-year self-neglect incidence rate is 241 (8.4%). A 10-item predictive model (with a c-statistic of 0.76) was developed using stepwise selection in multivariable logistical regression models. After corrections of overfitting by validating in 100 bootstrapping samples, the predictive accuracy of the model dropped to 0.71, suggesting at least moderate overfitting. A point-based risk index was developed based on parameter estimates of each predictive factor in the final logistic model. The index has an area under the receiver operating characteristic curve of 0.76.

CONCLUSION:

The study developed an efficient index with good predictive ability of self-neglect. Further external validation and impact studies are necessary before practitioners can apply this index to determine risk of self-neglect among other community aging populations.

Self-neglect is the behavior of an older person that threatens his/her own health and safety or fails to provide himself/herself with adequate food, water, clothing, shelter, personal hygiene, medication, and safety precautions.¹ Despite the wealth of existing knowledge about associated factors of prevalent self-neglect in sociodemographic, socioeconomic, health behaviors, medical, cognitive, physical, social, and psychological aspects,^2–9 the etiology of incident self-neglect is largely unknown.^5,10–13 A lack of systematic and longitudinal examination of antecedents of self-neglect has hindered the prevention and intervention development.

Self-neglect is pervasive and has been linked to substantial mortality and morbidity.^2,14–18 Prevalence estimates of self-neglect among older adults are 22.8%,¹⁹ 21.7%,¹⁵ and 29.1%² in different community-dwelling populations. According to the US Adult Protective Services, self-neglect has been the most commonly reported (31%) and substantiated (37%) among all elder abuse cases,¹⁴ and 45% of self-neglecting people were older than 80 years.²⁰ Practitioners might encounter more and older patients/clients who are at the risk of self-neglect given the rapidly growing aging population.

However, detection and treatment of self-neglect have been difficult for the practitioners because of medical comorbidities (eg, depression and dementia)⁵ as well as respecting the rights to self-determination. Self-neglecting individuals are likely to remain undetected until the conditions become so severe that they necessitate the involvement of healthcare and/or social service systems. The self-neglect cases encountered by physicians in clinical settings often have become so socially and medically intricate that extensive personnel and resources are required.²¹

Rigorously developed prognostic indices that systematically considered key factors are crucial in improving patient outcomes.^22,23 Because of the detection and treatment difficulties, a tool for prognosis regarding incident self-neglect is particularly necessary to optimize care, prevent, and intervene early. Healthcare professionals are well situated to assess the risk for self-neglect; a study of Adult Protective Services–identified older persons who self-neglect suggests that 92% of them have visited a primary care physician recently.⁹ But, current measurements of self-neglect are unable to provide prognosis and require in-home environmental observations,^7,24,25 which are often difficult and unfeasible to the practitioners.

Our goals are to (1) examine predictive factors of self-neglect onset and (2) develop a simple point-based predictive index to assess vulnerability prognostication of self-neglect among community-dwelling older adults.

METHODS

Study Populations

The Chicago Health and Aging Project (CHAP) is a prospective, population-based study that was aimed to identify risk factors of common health outcomes, particularly Alzheimer disease, among three biracial neighborhoods of the south side of Chicago, IL. Eligibility criteria for the project were non-Hispanic black or non-Hispanic white community-living individuals aged 65 years and older in the three adjacent neighborhoods. Hispanic neighborhoods were not considered to maximize power of the comparisons.

The present study used two waves of data from 2885 participants who completed both times of data collection during 2008 to 2012. The follow-up intervals were approximately 3 years after the prior wave. The predominant racial groups in CHAP were non-Hispanic blacks and whites. With informed written consent, the participants were interviewed by research assistants in private settings. Of the interviews, 75(2.6%) were proxy interviews. All interviewers went through standardized training. The study was approved by the Institutional Review Board of the Rush University Medical Center.^26,27

Dependent Variable: Incident Self-Neglect Onset

The outcome was incident self-neglect, defined by developing any new cases of self-neglect at the follow-up interviews (time 2) compared against the prior interviews (time 1). At each interview, participants were evaluated for 18 items, assessing four phenotypes of self-neglect: (1) hoarding, (2) poor personal hygiene, (3) unsanitary conditions, and (4) need home repair (Supplementary Table S1). Interviewers rated on a 4-point scale (0 = none, 1 = mild, 2 = moderate, and 3 = severe) for each item to judge a total score of self-neglect at both times. We used a common “any positive item” criterion to define self-neglect. Individuals who had none at time 1 and had mild or greater on any items at time 2 were considered as incident self-neglect onset. The Cronbach α of the self-neglect measurements was .95. Content, convergent, and predictive validity were found to be satisfactory.^15,17

Independent Variable: Considered Potential Predictive Factors of Self-Neglect

We tested 86 available variables from a list of nine predictive domains, and detailed instrument information is listed in Supplementary Text S1. The domains are: (a) sociodemographic/socioeconomic; (b) general well-being; (c) health behavior; (d) medical health; (e) medicine/healthcare; (f) cognitive function; (g) physical well-being; (h) social well-being; and (i) psychological well-being.

Analytic Approach

Descriptive analyses (means, SDs, and proportions) characterized study participants. For user convenience in practice, we created dichotomous variables for 86 considered predictive factors using multiple cutoff points based on the distributions (eg, age cutoffs: <65/≥65, <70/≥70, <75/≥75, <80/≥80, and <85/≥85 years). Each variable was tested for its bivariate association with self-neglect; a pool of variables with P < .1 was forwarded to stepwise selection in multivariable logistical regression models. The stepwise selection determines a final model consists of variables that were independently associated with incident self-neglect with P < .05.

We evaluated accuracy of the model by discrimination and calibration of the model.^22,28 Discrimination is often measured by the Harrell’s c-statistic, representing how well the model separates subjects who did and did not have incident self-neglect by estimated risk of the outcome. A c-statistic of 0.70 to 0.79 is considered as good.²² Calibration is the extent to which the predicted risk of the outcome matches the actual risk of the outcome. The Hosmer-Lemeshow (HL) P value is the most common measure of calibration.

To validate the models, we used the bootstrap algorithm presented by Harrell and colleagues in 1996,²⁹ estimating a correction for overfitting (“optimism”) of predictive models. The validation began by drawing 100 repeated bootstraps from the original sample; each has the original sizes of the samples. In each of the boot strap samples, we repeated the stepwise regression described above to fit 100 models with 100 c-statistics (Cboot). We then estimate another 100 c-statistics (Cbo) by applying the fitted model from boots trap data to original data. A measure of optimism was calculated by averaging the Cboot minus Cbo. Last, the “optimism-corrected” accuracy was evaluated by the original c-statistic minus the optimism.

Furthermore, we assessed the repeatability of each candidate variable based on its frequency of being selected as an independent predictor of self-neglect among the 100 bootstrap samples.³⁰ This method provides an estimation of strength of that variable to be repeatedly included in a multivariable model for new samples derived from the same populations.

To enable the calculation of risks of self-neglect based on the final logistic regression model, we constructed a point-based predictive index of self-neglect. Each predictive factor was assigned a score by dividing its own parameter estimates (${\widehat{b}}_i$) by the lowest ${\widehat{b}}_i$ in the model and rounded to the closet integer. The sum of scores of affirmative answers to the index items would be an individual’s predicted risk score of self-neglect. To measure the accuracy of the simplified index, we used the receiver operating characteristic (ROC) curves to contrast the true-positive rates (sensitivity) against the false-positive rate (1 – specificity). All statistical analyses were conducted using SAS, Version 9.2 (SAS Institute Inc).

For additional interpretation, each risk score was converted to a probability (0%−100%) of self-neglect. The process started with averaging the sum of intercept and all ${\widehat{b}}_i$ by the number of items in the final logistic regression model. Starting with the intercept value (score = 0), one-point increase in the risk score is associated with one-unit increase by the averaged value. A logistic number (n) was calculated corresponding to each risk score by the process. We then converted each logistic number (n) into a logistic probability (0%−100%) of having self-neglect using “exp (n)/[1+exp (n)]*100%”.

RESULTS

Descriptive Statistics and Incidence of Self-Neglect

Among the 2885 participants, 241 (8.4%) had incident self-neglect cases over the 3-year follow-up period. The mean (±SD) age of all participants was 76.8 (±6.7) years. A total of 1898 (65.8%) were women, and 1786 (61.9%) were black. The mean (±SD) of completed years of education was 13.2 (±3.3). For personal annual income (US dollars), 467 (16.3%) were within $0 to $14 999, 1093 (38.1%) were within $15 000 to 29 999, and 1306 (45.6%) reported more than $30 000. A total of 1157 (40.1%) were married, and 1728 (59.9%) were not married. For self-reported general health, 91 (3.2%) reported poor, 613 (21.3%) reported fair, 1507 (52.2%) reported good, and 674 (23.4%) reported excellent. For medical comorbidities, 535 (18.5%) reported none, 1287 (44.6%) reported one, 758 (26.3%) reported two, and 305 (10.6%) reported more than three.

Predictive Models of Self-Neglect, Model Performance, and Validation Using 100 Bootstrap Samples

Tables 1–3 show dichotomous variables that were significant predictors of incident self-neglect at the P < .1 level in bivariate analysis. These variables entered multivariable logistic regression analyses using stepwise selection, which resulted in a 10-item model. The model has a c-statistic of 0.76 and an HL P value of .91 (Table 4). Table 4 also presents the validation results using 100 bootstrap samples drawn from the original data set. The c-statistic dropped to 0.71 after the validation.

Table 1.

Potential Predictive Factors With Significant Bivariate Associations With Incident Self-Neglect at P < .1 Level

Predictive Factors		UOR (90% CI)	P Value
Sociodemographic/Socioeconomic
Age, y	≤75	1.30 (0.59–1.00)	.05
	≤80	1.37 (0.54–1.00)	.05
Race	African American	8.07 (5.01–12.97)	<.0001
Income, US $	<10 000	2.10 (1.38–3.19)	.00
	<15 000	2.18 (1.61–2.95)	<.0001
	<20 000	2.38 (1.81–3.11)	<.0001
	<25 000	2.50 (1.90–3.28)	<.0001
	<30 000	2.44 (1.81–3.28)	<.0001
Education, y	<8	1.75 (0.94–3.27)	.08
	<10	1.67 (1.14–2.44)	.01
	<12	1.85 (1.39–2.48)	<.0001
	<13	1.70 (1.30–2.24)	.00
	<15	1.92 (1.36–2.73)	.00
General Well-Being
Overall health (excellent/good/fair/poor)	Fair or poor	1.61 (1.21–2.13)	.00
Pain interference, d	>1	1.42 (1.09–1.85)	.01
Health Behavior (Yes/No)
Smoking	Current smoker (yes)	1.73 (1.13–2.63)	.01
Drinking	Alcohol (yes)	0.32 (0.23–0.47)	<.0001
	Wine (yes)	0.28 (0.17–0.44)	<.0001
	Beer (yes)	0.52 (0.32–0.83)	.01
	Liquor (yes)	0.37 (0.22–0.63)	.00
Medical Health (Yes/No)
Hypertension	SBP ≥150 mm Hg/DBP ≥90 mm Hg	2.12 (1.62–2.78)	<.0001
Diabetes	Yes	1.79 (1.20–2.67)	.00
Hypertension	Yes (self-reported)	1.38 (1.01–1.89)	.04
Shortness of breath	Yes	1.99 (1.51–2.61)	<.0001
Shortness of breath lying down	Yes	1.97(1.20–3.23)	.01
Knee pain	Yes	1.28 (0.97–1.69)	.08
Joint stiffness	Yes	1.48 (1.12–1.94)	.01
Medicine/Healthcare (Yes/No)
Estrogens	Yes	0.45 (0.19–1.01)	.05
Nursing home	Yes	0.30 (0.07–1.23)	.09
Aspirin	Yes	0.72 (0.56–0.94)	.02
Health supplement	Yes	0.64 (0.49–0.84)	.00
Nonprescribed medication	Yes	0.78 (0.59–1.01)	.07
No. of medicines in past 2 wk	>0	0.57 (0.33–0.98)	.04
	≥2	0.65 (0.43–0.98)	.04

Abbreviations: CI, confidence interval; DBP, diastolic blood pressure; SBP, systolic blood pressure; UOR, unadjusted odds ratio.

Table 3.

Potential Predictive Factors With Significant Bivariate Associations With Incident Self-Neglect at P < .1 Level

Predictive Factors		UOR (90% CI)	P Value
Social Well-Being
Social engagement (5 = everyday to 1 = once a year or less)	Religious service <5	3.37 (1.24–9.18)	.02
	Religious service <4	1.38 (1.05–1.81)	.02
	Visit museum <3	1.80 (1.20–2.70)	.01
	Go to movies <5	1.65 (1.10–2.47)	.02
	Go to movies <4	2.42 (1.84–3.18)	<.0001
	Go to movies <3	2.09 (1.56–2.79)	<.0001
	Visit relatives, friends, or neighbors <5	1.93 (1.36–2.75)	.00
	Visit relatives, friends, or neighbors <4	2.01 (1.54–2.62)	<.0001
	Visit relatives, friends, or neighbors <3	2.24 (1.70–2.95)	<.0001
	Have dinner or party <4	1.90 (1.39–2.58)	<.0001
	Have dinner or party <3	2.09 (1.61–2.73)	<.0001
	Go on trips <4	1.69 (0.99–2.90)	.05
	Go on trips <3	2.38 (1.79–3.16)	<.0001
	No. of museum visits past 10 y >0	0.41 (0.32–0.54)	<.0001
	No. of museum visits past 10 y ≥2	0.40 (0.29–0.54)	<.0001
	No. of museum visits past 10 y ≥3	0.38 (0.25–0.58)	<.0001
	No. of museum visits past 10 y ≥4	0.48 (0.28–0.80)	.01
	No. of concerts past 10 y >0	0.63 (0.47–0.83)	.00
	No. of concerts past 10 y ≥2	0.56 (0.43–0.74)	<.0001
	No. of concerts past 10 y ≥3	0.51 (0.37–0.70)	<.0001
	No. of concerts past 10 y ≥4	0.52 (0.35–0.76)	.00
	No. of library visits past 10 y >0	0.52 (0.40–0.68)	<.00
	No. of library visits past 10 y ≥2	0.51 (0.38–0.67)	<.0001
	No. of library visits past 10 y ≥3	0.38 (0.27–0.55)	<.0001
	No. of library visits past 10 y ≥4	0.38 (0.25–0.58)	<.0001
Psychological Well-Being
Yes/no	Felt depressed (yes)	1.80 (1.30–2.50)	.00
	Was happy (yes)	0.69 (0.47–1.01)	.05
	People were unfriendly (yes)	2.07 (1.35–3.17)	.00
	People disliked me (yes)	2.73 (1.39–5.34)	.00
	Treated with less courtesy (yes)	0.70 (0.52–0.93)	.02
	Feel God’s presence (yes)	2.07 (1.11–3.84)	.02
	Find comfort in religion (yes)	2.20 (1.02–4.73)	.04
	Feel God’s love (yes)	1.84 (1.01–3.33)	.05
	Desire to be closer with God (yes)	1.85 (1.02–3.35)	.04
	Pray or mediate (yes)	1.72 (0.92–3.20)	.09
Scale (4 = often to 1 = never)	Received poorer service 4 or 3)	0.74 (0.52–1.04)	.08
	Were called names or insulted (4 or 3)	1.71 (1.13–2.59)	.01
	Unable to control important things (4 or 3)	1.47 (1.10–1.95)	.01
	Things were not going your way (4 or 3)	1.43 (1.09–1.86)	.01
	Difficulties were piling up high (4 or 3)	1.90 (1.42–2.55)	<.0001
Scale 5 = strongly to 1 = strongly disagree)	Seldom sad (5 or 4)	1.43 (1.09–1.86)	.01
	Feel worthless (5 or 4)	1.55 (0.94–2.56)	.08
	Discouraged and giving up (5 or 4)	1.37 (0.96–1.95)	.08

Abbreviations: CI, confidence interval; UOR, unadjusted odds ratio.

Table 4.

The 10-Item Predictive Vulnerability Model for Self-Neglect, Model Performance, and Validation in 100 Bootstrap Samples Drawn From the Original CHAP Data Set

Predictive Domain	Predictive Factors	AOR (95% CI)	P Value
Sociodemographic/socioeconomic	African American	4.82 (2.94–7.93)	<.0001
	Ages ≤75 y	1.44 (1.08–1.92)	.01
	Income <$15 000	1.55 (1.15–2.09)	.00
Medical health	Hypertension	1.66 (1.25–2.21)	.00
	Shortness of breath	1.42 (1.06–1.90)	.02
Healthcare	Have not taken aspirin in past 3 y	1.39 (1.05–1.84)	.02
Physical well-being	Physical activity time <2 h in past 2 wk	1.45 (1.04–2.02)	.03
	Poor vision	2.06 (1.43–2.97)	<.0001
	Wear hearing aid	2.24 (1.02–4.94)	.05
Social engagement	Group dinner and party less frequent than several times a month	1.54 (1.11–2.14)	.01

Note: The c-statistic was 0.76, the Hosmer-Lemeshow P value was .91, and the optimism-corrected c-statistic was 0.71.

Abbreviations: CI, confidence interval; OR, adjusted odds ratio.

Repeatability of Predictive Factors in Bootstrap Samples

Supplementary Table S2 summarizes the frequency of each variable from the 10-item model that was selected as an independent predictor of self-neglect by repeating the stepwise regression in 100 bootstrap samples drawn from the original data set. Three variables (black, poor vision, and hypertension) were selected as independent predictors of self-neglect in at least 82% of the bootstrap samples. Two variables (wearing hearing aid and shortness of breath) were selected in at least 47% of the bootstrap samples. Three variables (less group dinner, lower income, and less physical activity) were selected in a minority (19%−31%) of the bootstrap samples. Two additional variables (aged <75 years and no aspirin taken) were not selected as independent predictors of self-neglect in any bootstrap sample.

Point-Based Vulnerability Risk Index of Self-Neglect

Supplementary Table S3 presents a point-based predictive index of self-neglect. The index ranges from 0 to 17, depending on responses to each item. Figure 1 presents the ROC curves of the validated predictive model and the risk score index. The index has an area under the receiver operating characteristic curve of 0.76.

Figure 1.

Accuracy of the risk score index compared against the predictive model using receiver operating characteristic curves.

DISCUSSION

In a longitudinal study of 2885 community-dwelling older adults, we developed a 10-item predictive index of self-neglect that can be administrated within several minutes and does not require an in-home assessment. The index demonstrated good predictive accuracy among a biracial US population (black and white) by an optimism-corrected c-statistic of 0.71. With further external validations and model updates in different populations, this prognostic information could assist the determination of vulnerability and provide opportunities to optimize care for older adults who are at risk.

The findings suggest that having hypertension, shortness of breath, and no aspirin taken in the past three years were predictive of self-neglect. A previous cross-sectional study (N = 500) found that hypertension is the most prevalent (52%) medical diagnosis among older patients who self-neglect.³¹ Additionally, evidence exists to support the preventive effect of low-dose aspirin on cardiovascular disease.³² Another study found that self-neglect was associated with an increased risk of cardiovascular-related mortality.¹⁷ These findings indicate a key role that hypertension and cardiovascular disease might have played in self-neglect in older persons.

Extending previous cross-sectional evidence about the close relationship between social engagement and self-neglect,^6,9 our study found that less engagement of social activity by having a dinner party with friends and family less frequently was a predictive factor of self-neglect. The finding reinforced the importance for older adults to maintain active engagement in social activities. In particular, engagement with friends and family should be examined as a key component in future prevention research of self-neglect.

Poor vision and wearing a hearing aid are predictors of self-neglect. Without assistance, older adults with impaired sensory function might have difficulties in providing themselves enough care and needs. Previous studies have also found that hearing and/or vision impairments may lead to social isolation,^33,34 decreased social participation,³⁵ and psychological distress.^36,37 The combined effect of the conditions might further increase one’s likelihood of self-neglect. More relevant support, such as Medicare coverage for sensory aids, is needed for older adults with sensory impairment to maintain quality life and sustain social participation.

Race/ethnicity needs careful considerations when developing prognosis tools. On the one hand, eliminating race can decrease predictive accuracy and lead to less optimal care for minority patients. Prior research in prognosis lacks population diversity and representativeness; physicians might rely on indices developed from white-majority populations for minority patients. On the other hand, including race/ethnicity in prognosis might involve misuses that reinforce racism, discrimination, and health disparities. Ongoing discussions suggest the importance of including minorities in future prognostication studies and including race as a predictor to enhance the generalizability. Therefore, it is essential to establish rigorous guidance for practitioners about how to consider race/ethnicity appropriately regarding patient outcomes.^38,39

In our study, after adjusting for a substantial list of factors, race/ethnicity (black as opposed to white) remains as an independent predictor of five times higher risk of self-neglect. Given the growing evidence about institutionalized inequalities experienced by racial/ethnic minority aging populations,^40–42 we highly suspect that there are unmeasured social and contextual factors leading to disparities, including self-neglect outcomes, despite personal-level factors (e.g., household and behaviors). The results call for more gerontology research and practice to address inequalities that accumulate across the life course and to test strategies that may help overcome the disadvantages over the trajectories of minority aging populations.

Although psychological distress and cognitive impairment are commonly concurrent conditions of self-neglect, our results indicate that they are not independent predictors of self-neglect after adjusting for various factors. According to the only large-scale (N = 2161) longitudinal examination of self-neglect, to our knowledge,⁵ cognitive impairment and depressive symptoms were predictive of severe self-neglect cases corroborated by the state’s ombudsman among a mainly white population (white = 82%) in Connecticut. Different adjusted factors and populations might partially explain the different findings. It is also possible that self-neglect onset develops before the progression of clinically significant cognitive impairment and depressive symptoms, which further magnify one’s inability for self-care and result in severe self-neglect conditions. While meaningful, the causal pathways cannot be determined by current predictive relationships and warrant more investigations.

In producing and validating the first predictive index for self-neglect among older adults,our study has several strengths and implications. Future studies could validate and/or modify this 10-item index, which might yield efficient tools for practitioners and caregivers to estimate patients’ risk of self-neglect and optimize care early. Second, few prognostic indices for health outcomes among older adults have been validated,²² and the bootstrap validation method we adopted provides more unbiased estimates of predictive accuracy compared to data splitting and cross-validation.²⁹ Third, various key factors considered in this longitudinal examination provide a systematic understanding of the antecedents of self-neglect. Fourth, the findings add vital information to the current self-neglect sociodemographic and socioeconomic profile. Our study suggests that among adiverse older population, younger, nonwhite, and lower-income individuals have higher vulnerability for incident self-neglect.

This study has limitations. First, our model has not been validated externally, and its generalizability to other populations is unclear. The predictive accuracy of our model dropped from 0.76 to 0.71 after the internal validation, suggesting overfitting and the drop would likely be greater when applied to another population. Recalibration and model revisions using diverse population datasets are necessary to increase its utility.^43–45 The study also did not include Hispanic older adults; self-neglect needs to be examined in this population. Moreover, by considering a substantial list of potential predictive variables, our model might be subject to overinclusion. However, at the initial stage of index development, we believe that sensitivity is preferable to specificity (underinclusion) and it is better to exclude false-positive variables in future confirmatory studies. The predictive factors are also limited regarding their causal implications for self-neglect. More empirical evidence and theoretical explanations are necessary to illuminate the etiology and causes of self-neglect. Last, our study does not provide evidence regarding the benefits and harms of using a prognostic model to prevent self-neglect, which must be quantified through impact studies.⁴⁵

In conclusion, this study developed an index that has good predictive ability for self-neglect among a diverse community population. With proper validations and modifications in different populations, the prognostication information has the potential to inform one who provides care or service to an older person a prognosis of self-neglect, which might be critical for improved health outcomes of the older person. The findings uncovered complex social and medical facets regarding self-neglect among older adults, which requires multidisciplinary strategies to be examined in future experimental investigations for the best practice. Leveraged efforts and funding resources are needed to reinforce legislation addressing justice for all, especially among the diverse aging populations.

Supplementary Material

Supporting information

Supplementary Text S1: Potential Predictive Factors of Self-Neglect.

Supplementary Table S1: Self-Neglect Measurement.

Supplementary Table S2: Frequency of the 10 Items Selected Using Stepwise Selection in 100 Bootstrap Samples Drawn From the Original CHAP Data Set.

Supplementary Table S3: A Point-Based 3-Year Vulnerability Risk Index of Self-Neglect.

Table 2.

Potential Predictive Factors With Significant Bivariate Associations With Incident Self-Neglect at P < .1 Level

Predictive Factors		UOR (90% CI)	P Value
Cognitive Function
MMSE	Remember an object	1.54 (0.98–2.41)	.06
	Remember object “table”	1.33 (1.01–1.76)	.05
	Repeat a phrase	1.52 (1.13–2.03)	.00
	Fold a paper	0.64 (0.39–1.07)	.09
	Place a paper in lap	1.38 (0.94–2.03)	.10
	Copy a drawing	1.53 (1.16–2.03)	.00
Physical Well-Being (Yes/No)
Five-chair stand	No	1.65 (1.22–2.23)	.00
10-s Tandem stand	No	1.77 (1.36–2.30)	<.0001
Timed 3.8’ seconds walk	No	1.91 (1.18–3.09)	.01
Poor vision	Yes	2.07 (1.48–2.90)	<.0001
Wear hearing aid	Yes	3.92 (1.82–8.39)	.00
Walk across a room	Yes	1.55 (1.02–2.33)	.04
Bath or shower	Yes	1.52 (0.96–2.39)	.07
Shopping	Yes	1.39 (1.02–1.90)	.04
Personal grooming	Yes	1.32 (0.96–1.82)	.09
Do laundry	Yes	1.54 (1.12–2.11)	.01
Walk for exercise	Yes	0.57 (0.42–0.77)	.00
Gardening	Yes	0.68 (0.50–0.93)	.02
Calisthenics	Yes	0.53 (0.37–0.76)	.00
Sports hobbies	Yes	0.29 (0.15–0.56)	.00
Physical activity	0	1.90 (1.46–2.47)	<.0001
	<2	2.38 (1.74–3.24)	<.0001
	<3	2.43 (1.70–3.47)	<.0001
ADL	>0	1.31 (0.97–1.76)	.08
Physical performance battery	<6 Scores	1.50 (1.15–1.95)	.00
Social Well-Being
Social engagement (5 = everyday to 1 = once a year or less)	Listen to radio <5 Listen to radio <4	1.69 (1.29–2.22) 1.50 (1.14–1.97)	.0001 .00
	Listen to radio <3	1.60 (1.19–2.15)	.00
	Read newspaper <5	1.78 (1.36–2.31)	<.0001
	Read newspaper <4	1.80 (1.37–2.35)	<.0001
	Read newspaper <3	2.07 (1.56–2.75)	<.0001
	Read magazine <5	1.40 (0.98–1.99)	.06
	Read magazine <4	1.79 (1.35–2.36)	<.0001
	Read magazine <3	2.08 (1.59–2.72)	<.0001
	Read magazine <2	1.87 (1.38–2.54)	<.0001
	Play games <5	1.47 (1.03–2.09)	.03
	Play games <4	1.61 (1.18–2.18)	.00
	Play games <3	1.86 (1.41–2.46)	<.0001
	Play games <2	1.80 (1.38–2.35)	<.0001
Social network size, No.	<3	1.36 (0.98–1.89)	.06
	<6	1.29 (0.99–1.69)	.06

Abbreviations: ADL, activity of daily living; CI, confidence interval; MMSE, Mini-Mental State Examination; UOR, unadjusted odds ratio.