| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
Correspondence: Address correspondence to Mark Lachs, MD, MPH, Professor of Medicine, Division of Geriatric Medicine and Gerontology, The Weill Medical College of Cornell University, 1300 York Avenue, Box 39, New York, NY 10021. E-mail: mslachs{at}mail.med.cornell.edu
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
Key Words: Crime victimization • Geriatric syndromes • Long-term care • Nursing home placement
Recently our group has become interested in the epidemiology of crime experienced by older adults. In linking a longitudinal cohort of older adults followed for over a decade, that is, the New Haven Established Population for Epidemiologic Studies in the Elderly (EPESE) cohort, to police records in the same catchment area, we demonstrated that nearly one third of the cohort had police interaction over a 10-year follow-up period as victims, witnesses, complainants, and even offenders (Lachs, Bachman, & Williams, 2005). In this study we explored the hypothesis that exposure to crime victimization might make older people vulnerable to loss of independence and lead to an increased risk of nursing home placement because of physical or emotional frailty resulting from victimization. In addition, previous work has shown that when violent crime is committed against older people in the form of domestic elder abuse that is investigated and corroborated by adult protective services, the victim is at an increased risk of nursing home placement (Lachs, Williams, O'Brien, & Pillemer, 2002).
We chose nursing home placement because it is perhaps the most dramatic and undesirable outcome associated with functional decline from any exposure in older people, and because nursing home placement has been completely ascertained for those cohort members placed in Connecticut nursing homes. We believe this to be the first article to assess a health outcome of crime exposure in a large observational study of older people.
|
Methods |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
Because detailed police records were available from 1985 onward, for the purpose of this study we considered the EPESE interview conducted in 1985 as the baseline; individuals alive in that year (n = 2,321) formed the sample for this study.
At baseline, participants took part in a detailed interview covering broad medical, functional, demographic, and psychosocial domains. Researchers used standardized instruments to assess cognition, depressive symptomatology, social networks, sources of emotional and other support, and chronic conditions. Interviewers met with participants annually; they conducted the 1985, 1988, and 1994 interviews in person, whereas they conducted the 1986, 1987, 1989, and 1990 interviews by telephone.
Researchers conducted interrater reliability substudies to ensure data quality, and we assume mortality follow-up to be complete.
Identification of Cohort Members Placed in Long-Term-Care Facilities
We identified cohort members who were placed in long-term-care facilities for the purpose of "custodial care" through a linkage with the Connecticut Long Term Care Registry, an information system designed to ascertain placement in certified Connecticut nursing homes. We defined custodial nursing home care as the first nursing home placement from either the community or hospital in which the length of nursing home stay was at least 30 days; if a newly admitted nursing home resident required hospitalization within 30 days of nursing home admission, we considered the stay to be custodial if the patient was readmitted to the nursing home and had a combined nursing home stay exceeding 30 days by adding the length of stays of all nursing home admissions. Of note, during this secular period "subacute care" (the use of long-term-care facilities as an adjunct to hospital care with an ultimate community discharge plan) was not common; diagnosis-related groups, which limit the length of inpatient stay and hospital reimbursement, had only recently come into being.
Linkage With Police Department Records
Linkage to police records was an arduous process that we performed manually (Lachs et al., 2001). After approval of the institutional review boards at both the Yale School of Medicine and the Weill Medical College of Cornell University, we matched police records to cohort members. Although the majority of EPESE respondents were deceased at the time of the data match, we used an extensive confidentiality protocol to protect the identity of all participants. We manually matched the 2,321 participants against police reports for the years 1985–1995. We accomplished this by comparing personal identifiers for all members of the cohort with demographic information on all police activity reports over the 11-year period. Although the EPESE cohort was initiated in 1982, 1985 was the first year that complete police reports were still available at the time of the study; we excluded cohort members not alive in 1985 from this analysis. We considered cases to be a match if there was agreement on (a) name and date of birth, (b) name and address, or (c) name and phone number. We considered records that could not be matched unequivocally to be nonmatches, and we include them in the comparison group in the subsequent analyses.
Two of us conducted an interrater reliability study to determine whether there was undue variability in the matching protocol, because this exercise directly determined the composition of the two major groups that we compare in this article. We did the interrater matching on a random subsample of 200 participants. There was agreement on the presence or absence of a match for 189 of these 200 participants (94.5% agreement); this corresponded to a kappa statistic of 
= 0.87, indicating excellent concordance.
For respondents deemed to be matches, we created a detailed abstraction form, which included the cohort member's role in the police event (e.g., as victim, perpetrator, complainant, or witness), the crime type, the cohort member's relationship to the perpetrator(s), whether a weapon(s) was used, number of offenders, if there was physical contact between the victim and offender, and if the victim needed medical treatment. We considered a cohort member to a victim if he or she was the target of an event; a perpetrator if he or she was an actor in a crime event; a complainant when he or she reported an event to police; and a witness when he or she witnessed a crime event. In certain situations it was possible to have more than one role in a single crime event (e.g., a domestic dispute in which a cohort member could be both a victim and perpetrator). These were uncommon, but for the purposes of analyses a cohort member could fall into only one group for a given incident. Through a consensus process, we developed a decision rule wherein the role of victim always superseded another role. Accordingly, we conducted another interrater reliability study on a sample of abstracted records that included a range of violent and nonviolent events to determine the observer variability of the abstraction process for incident and subject or role type (n = 40). The results again indicated excellent agreement for all variables tested, including subject type ( = 0.81), incident type ( = 0.97), whether medical treatment was rendered after the episode ( = 1.0), and whether the victim had actual physical contact with the victim ( = 1.0).
Law enforcement databases typically categorize crimes as violent or nonviolent. Violent crimes need not result in direct physical injury; they also involve simply the threat of injury (e.g., robbery at gunpoint, threatened assault). Of the violent crimes experienced by cohort members in this study, 50.3% involved assault and 44.8% involved robbery. The remaining 4.9% were categorized as "other."
After abstraction, we linked each police record with its corresponding EPESE record. The final merged electronic record created a unified police–health dataset without individual participant identifiers.
Strategy of Analysis
The primary exposure (independent variable) in this research was the experience of crime victimization by the cohort member. For this study, we examined the effects of four binary victimization subcategories: whether the cohort member experienced any victimization, whether the victimization was perpetrated by a stranger, whether the victimization was a violent crime, and whether the victim was injured during victimization.
Covariates in this research were individual characteristics previously shown to be predictors of nursing home placement, because these could theoretically confound the relationship between crime victimization and nursing home placement if one existed (Branch & Jette, 1982; Foley & Ostfeld, 1992; Freedman, 1996; Green & Ondrich, 1990). We measured each participant's functional status by using both the number of basic activity of daily living (ADL) impairments (0–7)(Katz, Ford, Moskowitz, Jackson, & Jaffe, 1963) and the number of higher functional impairments (0–5)(Nagi, 1976). We measured cognitive impairment by using the participant's score on the Mini-Mental State Exam (MMSE)(Folstein, Folstein, & McHugh, 1975). We examined social network variables, which included whether the cohort member was married or was employed, and a social ties index, which indicated a cohort member's number of social ties, including group participation, church attendance, and monthly contacts with close friends and relatives. Researchers updated each of these, except the MMSE score and social ties, which were measured only in the face-to-face interviews, for each wave of data collection. In the infrequent situation when surviving cohort members missed an interview or an item was skipped, values were carried forward from previous completed interviews. This was done for an average of 5.5% of participants across the six follow-up waves of data collection. Basic non-time-varying demographic covariates controlled for included age, gender, race, and education level. Because previous nursing home placement might also affect future nursing home placement, we controlled for this in the final model.
We used growth curve modeling, an analytic technique that has advanced tremendously over the past two decades, to measure change in risk over time. This technique can address analytic issues of longitudinal data that previous research using other statistical methods could not accommodate (Raudenbush & Bryk, 2002). Specifically, the technique permits the explicit modeling of individual change, while simultaneously allowing means and variances to vary over time by using many different time points. For the purposes of this article, individual growth curve analysis allows us to explain change in the risk of nursing home placement over multiple waves of data. Whereas aggregate-level techniques can estimate only a single slope for the entire sample as a representation of change over time in the probability of nursing home placement, with individual growth curve analysis, a single slope is estimated for each EPESE cohort member within the sample. This single slope is estimated from all observations simultaneously. The individual slope parameters then become the dependent variables in further analyses aimed at studying the effects of victimization and other medical and psychosocial variables over time, and accounting for individual variation with regard to these effects.
Because our dependent variable is binary (1 = nursing home placement within the time period and 0 = no placement), we utilized the hierarchical generalized linear model (HGLM) (Raudenbush & Bryk, 2002). The analysis of individual growth curves is considered hierarchical in that the seven intervals of measurement of nursing home placement are nested within individuals. In the first stage, we modeled the change in the log odds of nursing home placement across the entire seven time periods for each individual as follows (Level 1):
|
|
ij is the log odds of nursing home placement for individual i at time t; 
0i is the initial log odds of nursing home placement for individual i; 1i is the rate of change (slope) in the log odds of nursing home placement for individual i across the seven waves; and 
it represents random error in the measurement of nursing home placement for individual i at time t. Here, t refers to the timing, or year number, of the interviews at Wave 1 through Wave 7. During this first stage of analysis, we use these parameters to estimate the population average log odds of nursing home placement at baseline (1985), the average growth rate for this probability of placement across the seven waves for the entire sample, and the degree to which respondents' baseline probability and growth trajectories deviate from the population mean. In the second stage of analysis, we expanded the Level 2 models so that the parameters from the individual growth curves are modeled as a function of individual characteristics. In essence, the purpose of this stage of the analysis is to account for the between-subjects variation in the rates of change of the log odds of nursing home placement. For ease of presentation, we refer to as the "risk of nursing home placement" in the remainder of the article.
|
Results |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
|
Nursing Home Placement in the Cohort
Of the original 2,321 participants alive in 1985, 747 (32%) were placed in nursing homes for custodial care over the follow-up period (i.e., for an indefinite stay not involving rehabilitation or some other brief intervention with the goal of returning home). The mean length of nursing home stay was 413 days, with a standard deviation of 570 days, confirming that these were nursing home admissions for long-term care.
The initial HGLM analysis (Table 2) revealed the slope for time (or interview wave) to be positive and statistically significant (p <.001), indicating that the risk of nursing home placement increased over the seven waves of the data. This would be anticipated for a cohort of older adults who were community dwelling at baseline. In addition, the chi-square test indicated that there is significant variation among individual cohort members in the rate of change in the risk of nursing home placement; this warrants the inclusion of other explanatory variables in the model to determine which of these independently influence this risk.
|
|
|
|
Discussion |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
Several explanations, such as trauma caused by victimization itself, confounding by undetected or measured factors associated with both victimization and nursing home placement, and bias introduced in the conduct of this research, could underlie the mechanism of increased risk of nursing home placement in such individuals; we have attempted to summarize these in Table 5. The most obvious explanation may be that physical injuries sustained in victimization could create immobility or other impairment that directly produces the need for assistance with basic self-care. We do not believe this to be the case in most situations for two reasons. First, only one fourth of violent victimization events resulted in injury, and the risk of nursing home placement associated with these events was no greater than for violent victimization generally (see Table 3). Second, most crimes categorized as violent do not involve actual physical contact between assailant and victim (i.e., a crime is categorized as violent even if injury is only threatened, such as robbery at gunpoint). Harrowing examples of "noncontact violent crime" included situations such as a victim chased unsuccessfully on foot in an attempted robbery, or objects hurled at a victim that did not actually strike him or her.
|
Strengths of this work include the large size of the dataset, its longitudinal nature and long follow-up, and the measurement of covariates by use of standardized instruments. It also represents a unique melding of two different datasets: one, criminal justice; the other, health. However, there are also limitations to this work. Although the number of victimization events in the cohort was large, the number of violent victimizations in the cohort was relatively small. Many cases of violent crime are never reported to police (Hart & Rennison, 2003), which means that it is possible for victimized older adults to be erroneously included in the comparison group. However, the direction of this bias would be to underestimate the influence of crime victimization on nursing home placement. Despite this limitation, we believe this to be the largest systematic exploration of this topic ever undertaken, and we believe it is meritorious of further study.
What is most important is that we emphasize that the relationship between crime and subsequent nursing home placement in older people identified in this research may not be causal. It is entirely plausible that crime victimization is simply a marker for individuals destined to be at higher risk for subsequent nursing home placement. However, this would still be an immensely important finding, in that it would represent yet another risk factor for losing community independence that could inform subsequent research and clinical practice for clinicians and nonclinicians alike. Research should be directed at untangling the mechanism and pathophysiology of this relationship in ways that are not possible in this study, and it should also focus on the development of aggressive multipronged interventions to be tested in older people who experience violent crime and appear to be at risk of losing their independence. Physicians who care for older people should be mindful that patients who experience violent crime are at such risk and assiduously address other medical problems that can contribute to functional decline. Nonmedical providers—such as law enforcement, community social workers, meals-on-wheels programs, and elder social service agencies—should also be especially attentive to the life course of citizen-clients who have recently experienced violent crime, because augmented services might avert or forestall the need for subsequent nursing home placement.
|
Footnotes |
|---|
1 Division of Geriatric Medicine and Gerontology, Cornell University, Ithaca, NY.
2 Department of Sociology and Criminal Justice, University of Delaware, Newark.
3 Cecil G. Sheps Center for Health Services Research, University of North Carolina, Chapel Hill.
4 Program on Aging, Yale University, New Haven, CT.
5 New Haven Police Department, New Haven CT.
Decision Editor: Linda S. Noelker, PhD
Received for publication December 15, 2005. Accepted for publication May 24, 2006.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  R. Bachman and M. L. Meloy The Epidemiology of Violence Against the Elderly: Implications for Primary and Secondary Prevention Journal of Contemporary Criminal Justice, May 1, 2008; 24(2): 186 - 197. [Abstract] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
