- Email: [email protected]
Locating Study Subjects
Locating Study Subjects: Predictors and Successful Search Strategies with Inmates Released from a U.S. County Jail Enrique Menendez, MD, Mary Castle White, MPH, PhD, and Jacqueline P. Tulsky, MD Department of Community Health Systems, School of Nursing, University of California, San Francisco, California (E.M., M.C.W.) and AIDS Division, Internal Medicine, University of California, San Francisco, California (J.P.T.)
ABSTRACT: Minimizing loss to follow-up in longitudinal studies is critical. The purpose of this study was to examine the ability to locate subjects recently released from jail, identify predictors of being able to find a subject, and describe effective search strategies for this unique population. The sample for this cohort study included study subjects who were sought for interview after release from jail. Inmates in the San Francisco City and County Jail were enrolled in a randomized trial of incentives to improve follow-up for tuberculosis therapy after release from jail. Sociodemographic, health-related, and extensive locating information was collected during baseline interviews in jail. The main outcome was successful location of the subject. Study personnel recorded data on the number and nature of attempts made to find subjects in order to describe successful search strategies. Of 254 persons sought for the postrelease interview, 188 (74.0%) were found. Primary English speakers were more likely than Spanish speakers to be found (relative risk: 3.2, 95% confidence interval: 1.5–6.7, p ⫽ 0.002). Nearly one quarter of subjects (24%) were found back in jail, and the remainder were found in the community. Phone calls and letters to the subjects, and personal contacts to family and friends were successful strategies for 53% of the subjects. Seeking persons in programs, such as shelters and drug and alcohol programs, was successful in finding 18% of English-speaking subjects. Outreach efforts in sections of the city where Latinos spent time, including popular restaurants and community gathering places, were successful in finding 13% of Spanish-speaking subjects. We conclude that study subjects released from jails can be successfully located using well-defined search protocols tailored to the ethnicity of the sample and including a variety of strategies. Employment of bilingual personnel is important when a large proportion of subjects is monolingual and non-English speaking. Control Clin Trials 2001;22:238–247 © Elsevier Science Inc. 2001 KEY WORDS: Jail, corrections, attrition, longitudinal study methods, locating study subjects, follow-up
Address reprint requests to: Dr. M.C. White, Associate Professor, Department of Community Health Systems, School of Nursing, University of California, San Francisco, 2 Kirkham St., San Francisco, CA 94143-0608 ([email protected]) Received October 10, 2000; accepted February 20, 2001 Controlled Clinical Trials 22:238–247 (2001) © Elsevier Science Inc. 2001 655 Avenue of the Americas, New York, NY 10010
0197-2456/01/$–see front matter PII S0197-2456(01)00133-7
Locating Study Subjects
239
INTRODUCTION Loss to follow-up can be a critical limiting factor in study design, analysis, and interpretation of longitudinal study data, resulting in reduced power and the potential for biased findings [1–3]. Characteristics of subjects associated with high loss to follow-up include alcoholism, substance abuse, and being from a marginalized population, such as homeless persons [4–9]. While some studies have focused on specific marginalized populations, none have described incarcerated populations who are released from custody to the general community. The population of inmates is particularly challenging, because characteristics of the inmates themselves and the nature of incarceration may influence success in follow-up. For example, Hispanics comprised 15.5% of jail inmates in 1999, and were 2.3 times as likely to be held in jail as were white non-Hispanics [10]. Inmates in general and Latinos in particular may have additional characteristics that could impact follow-up, such as language, acculturation, and immigration status. To minimize the loss of study subjects, research personnel have developed a number of strategies to find subjects in the community [1, 3, 11–13]. These strategies can collectively be called the search protocol, defined as the group of efforts aimed at finding study subjects for follow-up, to minimize attrition of a study sample, continue therapeutic or diagnostic protocols in clinical studies, and measure outcomes. McKenzie et al., in a review of research to date, synthesized key elements important in follow-up of marginalized populations [14]. They were: collection of contact information; thorough organization and documentation of efforts to locate subjects; staff training and support; use of phone, mail, and incentives; establishing rapport; assurance of confidentiality; contacting agencies to help locate persons; fieldwork; and attention to safety concerns. Characteristics of the study team and the protocol have also been identified as important in successful location of subjects for follow-up. Areán and Gallagher-Thompson suggested that cultural competence is the most important issue in studies of an ethnic minority population, including attention to the setting and to the language and culture of the staff administering study protocols [15]. The need to examine effective strategies to ensure retention of study subjects is critical, particularly among subjects from the population of incarcerated persons. The purpose of this study, as part of a larger clinical trial on adherence to therapy for latent tuberculosis (TB) infection, was to examine the ability to locate study subjects recently released from jail, identify predictors of being able to find subjects, and describe effective search protocols for this unique population.
METHODS Design, Setting, and Sample A prospective cohort design was used in this study of inmates in the San Francisco City and County Jail. The sample included inmates enrolled in the Tuberculosis Prevention Project, a clinical trial of interventions to improve completion of therapy for latent TB infection after release. The trial occurred between 1997 and 2000 (manuscript in review). The protocol tested education versus the promise of a financial incentive, on two outcomes: first visit to the
240
E. Menendez et al.
County TB clinic within 1 month of release and completion of therapy for latent TB infection. Inclusion criteria were English or Spanish language, agreement to be in the study, and treatment for latent TB infection. Excluded were inmates who were determined by the Sheriff’s personnel to be violent or to have serious psychiatric illness. Inmates in jail under care for human immunodeficiency virus (HIV) or AIDS were excluded because they were in special programs for follow-up after release. Study subjects were informed at enrollment that they would be contacted for a follow-up interview after release. As part of the trial protocol, attempts were made to interview subjects at the time of the first TB clinic visit. The sample for this analysis includes those who were sought for this follow-up interview in the community, including those who came to the TB clinic but did not remain long enough for study personnel to meet with them and those who did not go to the TB clinic at all. All study procedures, including consents, the format for gathering contact information, and releases to contact agency personnel, were approved by the University of California, San Francisco, Institutional Review (Human Subjects) Board. Data and Data Collection Independent Variables Sociodemographic and lifestyle variables were collected at the baseline interview, which was conducted in English or Spanish according to the subject’s preference. Information was obtained by self-report, and questions such as those regarding a perceived problem with drugs or alcohol were asked in general terms to protect inmates who were in trial when interviewed. Immigrant status and reason for incarceration were not available to us. Homelessness in the month before jail was determined by a positive response to the question “In the month before coming to jail, did you spend any night on the street or in a shelter, in other words, homeless?” Questions were also asked that would help the study team find the subjects after release. These questions regarded name, birth date, nickname, and Social Security number. Because inmates sometimes used another name during incarceration, time was spent asking about all other names used in the community. Other information collected included type of housing before jail and anticipated residence after release, along with addresses and phone and pager numbers. Data collected on contacts who might assist in locating the subjects fell into the three categories identified by Goldstein et al. [4]. They were: informal, or associations of high intimacy and immediacy such as family and friends; formal, or organizations of low intimacy such as landlords and places of employment; and control, such as corrections and public or private treatment programs. For informal contacts, subjects were asked to provide family or partner names, addresses, and phone numbers. Family members, in particular mothers, have been identified as persons who reliably maintained contact and got messages to clients who were lost to follow-up in a study of drug abusers [16]. Among formal networks, subjects were asked about shelters and other temporary living residences that might be used after release from jail. Questions were asked about General Assistance and other public assistance programs
Locating Study Subjects
241
and the locations where subjects reported picking up checks. Employment information included where the person worked and whether study personnel could attempt to contact the subject, or his or her employer, at the work setting. In addition, research personnel asked about places where subjects frequently “hang out,” such as restaurants, gyms, parks, or community settings where persons congregate to find day work. Among control networks, research assistants asked information about probation or parole officers and medical care settings such as drug and alcohol treatment programs. And finally, research personnel noted additional physical characteristics of the inmate and any other relevant information that the interviewer and the subject thought might be useful when attempting to locate him or her after release. Information to help in the search was collected for all study subjects in an interactive manner such that the subject could understand and see the information being recorded. Questions such as “If we tried all the contacts you gave us, and still couldn’t find you, if you were me, where would you try next?” were asked in a way that subjects were able to participate in thinking about their own search strategy. Subjects understood, as part of the consent process, that study staff would not mention TB or jail in any written or verbal communications with contacts or agencies. For the purpose of quantifying efforts to locate subjects, a search attempt was defined as an effort to find the subject based on the information obtained at enrollment into the clinical trial. These included a phone call; visit to a site named by the subject, such as a restaurant, home, or other community location; and the mailing of letters to subjects or contacts. General activities, such as regular contact with jail personnel to determine if the subject had returned to jail, were not counted when attempts were summed for each individual. Dependent Variable The study outcome was whether the subject was located, regardless of whether the interview for the clinical trial was completed. The subject was recorded as found if he or she was contacted in person or by phone, with confirmation of identifying information to validate identity, as agreed upon during the baseline interview. Search Protocol Information from each study subject was maintained on a spreadsheet that served as the mechanism to determine when subjects were released from jail and eligible for follow-up. Once a subject missed an interview at the TB clinic, study staff initiated the search protocol. The first efforts took place in the office and included phone calls to subjects or their contacts. If the data collected from the subject did not include phone numbers, letters were sent to the last known address, which has been shown to be an effective strategy [17]. If the subject had reported homelessness or did not give information indicating stable housing, shelters and other agencies were contacted on a weekly basis, and letters were provided to agency personnel to forward to the subject. Outreach efforts included visits to community sites, as provided in the baseline interview, to
242
E. Menendez et al.
look for the subject directly or to leave a letter with a person previously designated by the subject. After all attempts to find the subject were exhausted, research assistants continued to review jail records weekly in the event the subject was rearrested. Finally, the subject was considered lost if study personnel were unable to find him or her after 5 months of efforts (6 months after release from jail), using all available data and contacts. All search efforts were recorded on data collection sheets kept for each subject and updated weekly, including the type of effort, date, time, and result. Throughout the study 75% of personnel were fluent Spanish speakers, based on previous studies indicating that approximately 60% of the target population for the clinical trial were monolingual Spanish-speaking Latinos [18, 19]. The study team was trained and monitored for consistent use of the search protocol. Regular meetings with staff to review search efforts and solve problems served both to ensure that an adequate effort was made for each person and to maintain morale among the research personnel. Informal competition among study personnel, first reported by Craig as successful in maintaining enthusiasm, especially when progress was slow [16], was used as a positive tool during these meetings. The project director coordinated these meetings to reinforce training, monitor safety issues, and provide support to the tracking staff.
Data Analysis Subjects who were found were compared to those who were not found, by sociodemographic characteristics and factors associated with loss to follow-up such as drug or alcohol problems, social support, and housing stability. The sample was described using standard descriptive statistics. Bivariate comparisons were made, using t tests and chi-square analysis, to identify predictors of being found. Variables significant at alpha ⫽ 0.10 were entered into a logistic regression to identify significant predictors while controlling for covariates including study group from the clinical trial. Analyses were done using SPSS statistical software.
RESULTS Sample Characteristics and Predictors of Being Found The sample included 254 subjects, and of these 188 (74.0%) were found. They are described in Table 1 for all subjects who answered baseline questions about sociodemographic characteristics. Because of similar results in subcategories of marital status, this variable is presented and was analyzed further as a dichotomous variable, partnered (including married or partnered) and unpartnered (never married, separated, divorced, or widowed). The mean age of the sample was 30 years. On average, 5.7 attempts were made for those who were found (median ⫽ 5, range: 1–22), and 10.2 attempts were made for those who were not found (median ⫽ 9, range: 3–23). Over half of the subjects (57.8%) were Latino. Subjects who reported that they were Asian or Latino had the lowest rates of being located (64.3% and 66.0%, respectively) as compared to those reporting other ethnic backgrounds.
243
Locating Study Subjects
Persons who indicated a preference for English, versus Spanish, at baseline interview were 1.3 times as likely to be found (95% confidence interval: 1.2–1.5, p ⬍ 0.001). Place of birth (United States versus outside the United States) had nearly identical results as language preference and was highly correlated with this variable. Among subjects born outside the United States, longer time in the United States was associated with being found for interview. Time in the United States was categorized into three levels: 0–5 years, ⬎ 5 years but foreign-born, and U.S.-born subjects. This ordinal variable was significantly associated with being found, by the chi-square test for trend (p ⫽ 0.011) (Table 1). A Table 1 Characteristics of 254 Study Subjects Sought for Follow-up Interview after Release from Jail, by Whether or Not They Were Found Characteristic Gender Male Female Ethnicity Latino Black Other, mixed Asian White Native American Birth country United States Mexico Honduras El Salvador Other Latin America Asia/Pacific Islands Other Preferred language English Spanish Time in United States 0–5 years since immigration ⬎5 years since immigration Born in United States Partnership status Never married, separated, divorced, or widowed Married/partnered Housing before jail Ever homeless in last month Never homeless in last month Employed before jail Employed in last 3 months Unemployed in last 3 months Drugs or alcohol Perceived problem No perceived problem Total a
n
Found (n)
Found (%)
p-value
228 26
167 21
73.2 80.8
ns
147 54 26 14 11 2
97 47 23 9 10 2
66.0 87.0 88.5 64.3 90.9 100.0
0.001a
82 66 37 18 21 17 10
70 38 24 16 16 12 9
85.4 57.6 64.9 88.9 76.2 70.6 90.0
0.001a
109 145
94 94
86.2 64.8
⬍ 0.001
76 84 84
50 60 72
65.8 71.4 85.7
0.011b
167 87
128 60
76.7 69.0
ns
76 175
57 128
75.0 73.1
ns
144 110
109 79
75.7 71.8
ns
136 118 254
99 89 188
72.8 75.4 74.0
ns
Chi-square test of the value in this row as compared to all other values. Chi-square test for trend.
b
244
E. Menendez et al.
stepwise logistic regression analysis indicated that language (preferring English versus Spanish for baseline interview in jail) was the only variable that predicted being found (relative risk: 3.2, 95% confidence interval: 1.5–6.7, p ⫽ 0.002), while controlling for covariates and the clinical trial study group. Time to Locate Subjects and Effective Strategies In the subset of 188 persons who were located, the average time to location was 61.5 days (median ⫽ 36 days, range: 1–180). Of the 21 women found, on average the time to location was 45 days, as compared to 64 days for the 167 men who were located, although this difference was not statistically significant. This sample experienced a high rate of recidivism: overall, nearly one quarter of the subjects (24%) were found back in jail. Phone and personal contacts (family and friends) were the most successful strategies for 53% of the sample. Effective strategies differed among subgroups. Because we found that preferred language, ethnicity, and time in the United States were associated with being found, we described the strategies that were successful, stratified by these variables (Table 2). Data on successful strategies were missing for one subject. Seeking persons in shelters and drug and alcohol programs were successful strategies among non-Latinos and English-speaking subjects. Outreach efforts were successful, in particular among Latinos, Spanish-speaking subjects, and new immigrants. These efforts focused on sections of the city where ethnic groups spent time, including popular restaurants and community gathering places. DISCUSSION In a population of study subjects released from jail, we located 74.0% based on information gathered at baseline. Harlow reported successful tracking to
Table 2 Percent of Persons Found Using Different Search Strategies, by Characteristics Identified as Predictors of Being Found Predictor Language English (n ⫽ 93) Spanish (n ⫽ 94) Time in United States 0–5 years (n ⫽ 50) ⬎ 5 years (n ⫽ 60) Born in United States (n ⫽ 71) Ethnicity Latino (n ⫽ 97) Black (n ⫽ 47) Other, mixed (n ⫽ 22) White (n ⫽ 10) Asian, Native American (n ⫽ 11) Total (n ⫽ 187)
Jail
Phone/Letter
Contacts
Programs
Outreach
26 22
32 34
15 23
18 7
9 13
24 17 30
34 43 24
20 18 20
8 8 20
14 13 7
24 30 18 30 9 24
34 30 32 20 54 33
21 15 23 30 9 20
9 18 27 20 0 12
12 6 9 0 27 11
Contacts included individuals named by study subject. Programs included shelters, drug and alcohol programs, TB clinic, and other programs that subjects provided permission to contact. Outreach included visits to community settings, including popular restaurants and community gathering places.
Locating Study Subjects
245
completion of TB preventive therapy in 63% [20], and our rates of locating subjects were similar to those found by Ludwig et al. [21]. Results from this study focus on locating a specific group that has several differences from those studied by others. Released inmates are a group with special characteristics that may affect outcome. First, because this sample came from a study of TB in a west coast city, it included a disproportionate number of new immigrants, in particular from Latin America, and language was an important predictor of success in locating subjects after release. In addition to cultural differences, they may have had immigration issues that affected their willingness to be found for follow-up. Predictors of locating subjects were not in complete agreement with other research. Poor social functioning, including unpartnered or unmarried status, predicted loss to follow-up in a sample of 85 alcoholics [8]. Although not statistically significant, our findings indicated that persons who were married or partnered at baseline interview were less likely to be located after release from jail. In several instances, the subject named a partner as the main or only contact, and when approached, this contact person was no longer partnered with the former inmate. The fact of being in jail may have prompted the dissolution of the partnership, or, in the extreme, the reason for the person being in jail may have been related to the relationship. Senturia et al. found that the inability to provide several alternative contacts was predictive of retention problems [22]. This finding, combined with our results, leads us to believe that probing for as many contacts as possible is an important strategy, especially if the inmate provides only one partner as a contact. Our work also differed from that of MacKenzie et al. in the area of homelessness. In their study of alcoholics, residential instability in the follow-up period predicted loss to follow-up [8]. In our study, persons who reported homelessness before jail were more likely to be found, although the finding did not reach statistical significance. In this case as well, the fact of the jail stay may have influenced the ease with which persons were found after release. Persons with a history of homelessness were able to name shelters where they occasionally stayed, and these were stable locations as compared to persons who had one or two contact individuals and private residences that may have become disrupted as a result of their jail stay. Among effective strategies, jail itself has been identified as an important setting for locating subjects, in our study and in studies of persons with a history of time in jail. Whelan et al. found that among high-risk drug users in Oregon, 19% were found while they were incarcerated [23], which is similar to our rate of 24% who were found in jail. Ludwig et al. found that the most productive strategies in the community were mail and telephone contacts with relatives and friends [21]. In our sample of released inmates, known to have high rates of substance abuse and mental health problems, programs for drug and alcohol problems were successful places to search for subjects. And finally, in this predominantly Latino sample, our successful outreach efforts included questions about sections of the city where particular subgroups congregated for social or employment reasons. Sending staff to these areas was a successful strategy in particular for those who were new immigrants from Latin American countries. Staff preparation and cultural competence have been cited repeatedly in the
246
E. Menendez et al.
literature and were important in this study. Marmor et al. examined follow-up in the Framingham Children’s Study and found that attitudes of staff, feedback to subjects, and the staff’s handling of questions and problems were important predictors of continued participation [1]. They concluded that quality of the communication with study subjects and the subjects’ perceived importance of the research were key factors in minimizing loss to follow-up. In a study of drug abusers in St. Louis, Cottler et al. identified persistence, creative teamwork, and a comprehensive searching strategy as critical factors in successful follow-up [24]. Krieger et al. found that in a clinical trial in which enhanced searching and follow-up were parts of the intervention, community health workers of the same ethnicity and from the same communities may have accounted for the effectiveness of outreach [25]. Desland and Batey located 76% of heroin users for follow-up interviews, citing collection of good and reliable information, ethical practices in informing subjects of searching methods, confidentiality of the methods, and using creativity and flexibility in searching for subjects as reasons for their success [3]. In our study, research personnel were bilingual and bicultural, search protocols were consistent and monitored, and the study team was trained, monitored, and encouraged and maintained enthusiasm throughout the study. We believe these are critically important aspects of a study protocol, both to obtain sufficient contact information and to approach contacts appropriately and successfully in the community. Locating study subjects in longitudinal research is critical to achieve sufficient sample size for analysis and to make conclusions that can be considered valid. The population of released inmates shares characteristics, such as unstable housing, drug and alcohol problems, and unpartnered status, that have been identified as predictors of loss to follow-up. However, in this study these characteristics did not predict loss to follow-up, as incarceration per se seemed to influence the ability to locate study subjects. For this population, therefore, attention should be paid to collection of adequate residential and social contact information. Specific ethnic groups, such as Latinos in this study, differ in the social contacts and strategies that will be most successful. For those who prefer a language other than English, special efforts include employment of bicultural as well as bilingual personnel. Identification of those who are likely to be difficult to find can help project managers allocate resources appropriately. This project was funded by the National Institutes of Health, NINR (R01 NR04456).
REFERENCES 1. Marmor J, Oliveira S, Donahue R, et al. Factors encouraging cohort maintenance in a longitudinal study. J Clin Epidemiol 1991;44:531–535. 2. Freedman D, Thornton A, Camburn D. Maintaining response rates in longitudinal studies. Soc Methods Res 1980;9:87–98. 3. Desland M, Batey R. High retention rates within a prospective study of heroin users. Br J Addict 1991;86:859–865. 4. Goldstein PJ, Abbott W, Paige W, Sobel I, Soto F. Tracking procedures in follow-up studies of drug abusers. Am J Drug Alcohol Abuse 1977;4:21–30. 5. Bale R, Cabrera S, Brown J. Follow-up evaluation of drug abuse treatment. Am J Drug Alcohol Abuse 1977;4:233–249.
Locating Study Subjects
247
6. Moos R, Bliss F. Difficulty of follow-up and outcome of alcoholism treatment. J Stud Alcohol 1978;39:473–490. 7. Klonoff-Cohen H. Tracking strategies involving fourteen sources for locating a transient study sample: Parents of sudden infant death syndrome infants and control infants. Am J Epidemiol 1996;144:98–101. 8. MacKenzie A, Funderburk FR, Allen RP, Stefan R. The characteristics of alcoholics frequently lost to follow up. J Stud Alcohol 1987;48:119–123. 9. Pollio DE, Thompson SJ, North CS. Agency-based tracking of difficult-to-follow populations: Runaway and homeless youth programs in St. Louis, Missouri. Community Ment Health J 2000;36:247–258. 10. Beck AJ. Prison and jail inmates at midyear 1999. In: Bureau of Justice Statistics Bulletin. Washington, D.C.: U.S. Dept of Justice; April, 2000. NIJ 181643. 11. Wright J, Lampton A, Devine JA. Tracking non-traditional populations in longitudinal studies. Evaluation and Program Planning 1995;18:267–277. 12. Given B, Keilman L, Collins C, Given W. Strategies to minimize attrition in longitudinal studies. Nurs Res 1990;39:184–186. 13. Hough R, Tarke H, Renker V, Shields P, Glatstein J. Recruitment and retention of homeless mentally ill participants in research. J Consult Clin Psychol 1996;64:881–891. 14. McKenzie M, Peterson Tulsky J, Long HL, Chesney M, Moss A. Tracking and follow-up of marginalized populations: A review. J Health Care Poor Underserved 1999;10:409–429. 15. Areán PA, Gallagher-Thompson D. Issues and recommendations for the recruitment and retention of older ethnic minority adults into clinical research. J Consult Clin Psychol 1996;64:875–880. 16. Craig R. Locating drug addicts who have dropped out of treatment. Hosp Community Psychiatry 1979;30:402–404. 17. Eckland BK. Retrieving mobile cases in longitudinal surveys. Public Opinion Quart 1968;32:51–64. 18. Tulsky JP, White MC, Dawson C, et al. Screening for tuberculosis in jail and clinic follow-up after release. Am J Public Health 1998;88:223–226. 19. White MC, Tulsky JP, Reilly P, et al. A clinical trial of a financial incentive to go to the tuberculosis clinic for isoniazid after release from jail. Int J Tuberc Lung Dis 1998;2:506–512. 20. Harlow T. TB net tracking network provides continuity of care for mobile TB patients. Am J Public Health 1999;89:1581–1582. 21. Ludwig AM, Levine J, Stark LH. LSD and Alcoholism: A Clinical Study of Treatment Efficacy. Springfield, Illinois: Thomas, Inc.; 1970. 22. Senturia Y, Mortimer K, Baker D, Joseph C, Wedner J. Successful techniques for retention of study participants in an inner city population. Control Clin Trials 1998;19:544–554. 23. Whelan V, McBride D, Colby R. Public health department tracking of high-risk drug users. Public Health Rep 1993;108:643–645. 24. Cottler L, Compton W, Ben-Abdallah A, Horne M, Claverie D. Achieving 96.6 percent follow-up rate in a longitudinal study of drug abusers. Drug Alcohol Depend 1996;41:209–217. 25. Krieger J, Collier C, Song L, Martin D. Linking community based blood pressure measurement to clinical care: A randomized controlled trial of outreach and tracking by community health workers. Am J Public Health 1999;89:856–861.
ABSTRACT: Minimizing loss to follow-up in longitudinal studies is critical. The purpose of this study was to examine the ability to locate subjects recently released from jail, identify predictors of being able to find a subject, and describe effective search strategies for this unique population. The sample for this cohort study included study subjects who were sought for interview after release from jail. Inmates in the San Francisco City and County Jail were enrolled in a randomized trial of incentives to improve follow-up for tuberculosis therapy after release from jail. Sociodemographic, health-related, and extensive locating information was collected during baseline interviews in jail. The main outcome was successful location of the subject. Study personnel recorded data on the number and nature of attempts made to find subjects in order to describe successful search strategies. Of 254 persons sought for the postrelease interview, 188 (74.0%) were found. Primary English speakers were more likely than Spanish speakers to be found (relative risk: 3.2, 95% confidence interval: 1.5–6.7, p ⫽ 0.002). Nearly one quarter of subjects (24%) were found back in jail, and the remainder were found in the community. Phone calls and letters to the subjects, and personal contacts to family and friends were successful strategies for 53% of the subjects. Seeking persons in programs, such as shelters and drug and alcohol programs, was successful in finding 18% of English-speaking subjects. Outreach efforts in sections of the city where Latinos spent time, including popular restaurants and community gathering places, were successful in finding 13% of Spanish-speaking subjects. We conclude that study subjects released from jails can be successfully located using well-defined search protocols tailored to the ethnicity of the sample and including a variety of strategies. Employment of bilingual personnel is important when a large proportion of subjects is monolingual and non-English speaking. Control Clin Trials 2001;22:238–247 © Elsevier Science Inc. 2001 KEY WORDS: Jail, corrections, attrition, longitudinal study methods, locating study subjects, follow-up
Address reprint requests to: Dr. M.C. White, Associate Professor, Department of Community Health Systems, School of Nursing, University of California, San Francisco, 2 Kirkham St., San Francisco, CA 94143-0608 ([email protected]) Received October 10, 2000; accepted February 20, 2001 Controlled Clinical Trials 22:238–247 (2001) © Elsevier Science Inc. 2001 655 Avenue of the Americas, New York, NY 10010
0197-2456/01/$–see front matter PII S0197-2456(01)00133-7
Locating Study Subjects
239
INTRODUCTION Loss to follow-up can be a critical limiting factor in study design, analysis, and interpretation of longitudinal study data, resulting in reduced power and the potential for biased findings [1–3]. Characteristics of subjects associated with high loss to follow-up include alcoholism, substance abuse, and being from a marginalized population, such as homeless persons [4–9]. While some studies have focused on specific marginalized populations, none have described incarcerated populations who are released from custody to the general community. The population of inmates is particularly challenging, because characteristics of the inmates themselves and the nature of incarceration may influence success in follow-up. For example, Hispanics comprised 15.5% of jail inmates in 1999, and were 2.3 times as likely to be held in jail as were white non-Hispanics [10]. Inmates in general and Latinos in particular may have additional characteristics that could impact follow-up, such as language, acculturation, and immigration status. To minimize the loss of study subjects, research personnel have developed a number of strategies to find subjects in the community [1, 3, 11–13]. These strategies can collectively be called the search protocol, defined as the group of efforts aimed at finding study subjects for follow-up, to minimize attrition of a study sample, continue therapeutic or diagnostic protocols in clinical studies, and measure outcomes. McKenzie et al., in a review of research to date, synthesized key elements important in follow-up of marginalized populations [14]. They were: collection of contact information; thorough organization and documentation of efforts to locate subjects; staff training and support; use of phone, mail, and incentives; establishing rapport; assurance of confidentiality; contacting agencies to help locate persons; fieldwork; and attention to safety concerns. Characteristics of the study team and the protocol have also been identified as important in successful location of subjects for follow-up. Areán and Gallagher-Thompson suggested that cultural competence is the most important issue in studies of an ethnic minority population, including attention to the setting and to the language and culture of the staff administering study protocols [15]. The need to examine effective strategies to ensure retention of study subjects is critical, particularly among subjects from the population of incarcerated persons. The purpose of this study, as part of a larger clinical trial on adherence to therapy for latent tuberculosis (TB) infection, was to examine the ability to locate study subjects recently released from jail, identify predictors of being able to find subjects, and describe effective search protocols for this unique population.
METHODS Design, Setting, and Sample A prospective cohort design was used in this study of inmates in the San Francisco City and County Jail. The sample included inmates enrolled in the Tuberculosis Prevention Project, a clinical trial of interventions to improve completion of therapy for latent TB infection after release. The trial occurred between 1997 and 2000 (manuscript in review). The protocol tested education versus the promise of a financial incentive, on two outcomes: first visit to the
240
E. Menendez et al.
County TB clinic within 1 month of release and completion of therapy for latent TB infection. Inclusion criteria were English or Spanish language, agreement to be in the study, and treatment for latent TB infection. Excluded were inmates who were determined by the Sheriff’s personnel to be violent or to have serious psychiatric illness. Inmates in jail under care for human immunodeficiency virus (HIV) or AIDS were excluded because they were in special programs for follow-up after release. Study subjects were informed at enrollment that they would be contacted for a follow-up interview after release. As part of the trial protocol, attempts were made to interview subjects at the time of the first TB clinic visit. The sample for this analysis includes those who were sought for this follow-up interview in the community, including those who came to the TB clinic but did not remain long enough for study personnel to meet with them and those who did not go to the TB clinic at all. All study procedures, including consents, the format for gathering contact information, and releases to contact agency personnel, were approved by the University of California, San Francisco, Institutional Review (Human Subjects) Board. Data and Data Collection Independent Variables Sociodemographic and lifestyle variables were collected at the baseline interview, which was conducted in English or Spanish according to the subject’s preference. Information was obtained by self-report, and questions such as those regarding a perceived problem with drugs or alcohol were asked in general terms to protect inmates who were in trial when interviewed. Immigrant status and reason for incarceration were not available to us. Homelessness in the month before jail was determined by a positive response to the question “In the month before coming to jail, did you spend any night on the street or in a shelter, in other words, homeless?” Questions were also asked that would help the study team find the subjects after release. These questions regarded name, birth date, nickname, and Social Security number. Because inmates sometimes used another name during incarceration, time was spent asking about all other names used in the community. Other information collected included type of housing before jail and anticipated residence after release, along with addresses and phone and pager numbers. Data collected on contacts who might assist in locating the subjects fell into the three categories identified by Goldstein et al. [4]. They were: informal, or associations of high intimacy and immediacy such as family and friends; formal, or organizations of low intimacy such as landlords and places of employment; and control, such as corrections and public or private treatment programs. For informal contacts, subjects were asked to provide family or partner names, addresses, and phone numbers. Family members, in particular mothers, have been identified as persons who reliably maintained contact and got messages to clients who were lost to follow-up in a study of drug abusers [16]. Among formal networks, subjects were asked about shelters and other temporary living residences that might be used after release from jail. Questions were asked about General Assistance and other public assistance programs
Locating Study Subjects
241
and the locations where subjects reported picking up checks. Employment information included where the person worked and whether study personnel could attempt to contact the subject, or his or her employer, at the work setting. In addition, research personnel asked about places where subjects frequently “hang out,” such as restaurants, gyms, parks, or community settings where persons congregate to find day work. Among control networks, research assistants asked information about probation or parole officers and medical care settings such as drug and alcohol treatment programs. And finally, research personnel noted additional physical characteristics of the inmate and any other relevant information that the interviewer and the subject thought might be useful when attempting to locate him or her after release. Information to help in the search was collected for all study subjects in an interactive manner such that the subject could understand and see the information being recorded. Questions such as “If we tried all the contacts you gave us, and still couldn’t find you, if you were me, where would you try next?” were asked in a way that subjects were able to participate in thinking about their own search strategy. Subjects understood, as part of the consent process, that study staff would not mention TB or jail in any written or verbal communications with contacts or agencies. For the purpose of quantifying efforts to locate subjects, a search attempt was defined as an effort to find the subject based on the information obtained at enrollment into the clinical trial. These included a phone call; visit to a site named by the subject, such as a restaurant, home, or other community location; and the mailing of letters to subjects or contacts. General activities, such as regular contact with jail personnel to determine if the subject had returned to jail, were not counted when attempts were summed for each individual. Dependent Variable The study outcome was whether the subject was located, regardless of whether the interview for the clinical trial was completed. The subject was recorded as found if he or she was contacted in person or by phone, with confirmation of identifying information to validate identity, as agreed upon during the baseline interview. Search Protocol Information from each study subject was maintained on a spreadsheet that served as the mechanism to determine when subjects were released from jail and eligible for follow-up. Once a subject missed an interview at the TB clinic, study staff initiated the search protocol. The first efforts took place in the office and included phone calls to subjects or their contacts. If the data collected from the subject did not include phone numbers, letters were sent to the last known address, which has been shown to be an effective strategy [17]. If the subject had reported homelessness or did not give information indicating stable housing, shelters and other agencies were contacted on a weekly basis, and letters were provided to agency personnel to forward to the subject. Outreach efforts included visits to community sites, as provided in the baseline interview, to
242
E. Menendez et al.
look for the subject directly or to leave a letter with a person previously designated by the subject. After all attempts to find the subject were exhausted, research assistants continued to review jail records weekly in the event the subject was rearrested. Finally, the subject was considered lost if study personnel were unable to find him or her after 5 months of efforts (6 months after release from jail), using all available data and contacts. All search efforts were recorded on data collection sheets kept for each subject and updated weekly, including the type of effort, date, time, and result. Throughout the study 75% of personnel were fluent Spanish speakers, based on previous studies indicating that approximately 60% of the target population for the clinical trial were monolingual Spanish-speaking Latinos [18, 19]. The study team was trained and monitored for consistent use of the search protocol. Regular meetings with staff to review search efforts and solve problems served both to ensure that an adequate effort was made for each person and to maintain morale among the research personnel. Informal competition among study personnel, first reported by Craig as successful in maintaining enthusiasm, especially when progress was slow [16], was used as a positive tool during these meetings. The project director coordinated these meetings to reinforce training, monitor safety issues, and provide support to the tracking staff.
Data Analysis Subjects who were found were compared to those who were not found, by sociodemographic characteristics and factors associated with loss to follow-up such as drug or alcohol problems, social support, and housing stability. The sample was described using standard descriptive statistics. Bivariate comparisons were made, using t tests and chi-square analysis, to identify predictors of being found. Variables significant at alpha ⫽ 0.10 were entered into a logistic regression to identify significant predictors while controlling for covariates including study group from the clinical trial. Analyses were done using SPSS statistical software.
RESULTS Sample Characteristics and Predictors of Being Found The sample included 254 subjects, and of these 188 (74.0%) were found. They are described in Table 1 for all subjects who answered baseline questions about sociodemographic characteristics. Because of similar results in subcategories of marital status, this variable is presented and was analyzed further as a dichotomous variable, partnered (including married or partnered) and unpartnered (never married, separated, divorced, or widowed). The mean age of the sample was 30 years. On average, 5.7 attempts were made for those who were found (median ⫽ 5, range: 1–22), and 10.2 attempts were made for those who were not found (median ⫽ 9, range: 3–23). Over half of the subjects (57.8%) were Latino. Subjects who reported that they were Asian or Latino had the lowest rates of being located (64.3% and 66.0%, respectively) as compared to those reporting other ethnic backgrounds.
243
Locating Study Subjects
Persons who indicated a preference for English, versus Spanish, at baseline interview were 1.3 times as likely to be found (95% confidence interval: 1.2–1.5, p ⬍ 0.001). Place of birth (United States versus outside the United States) had nearly identical results as language preference and was highly correlated with this variable. Among subjects born outside the United States, longer time in the United States was associated with being found for interview. Time in the United States was categorized into three levels: 0–5 years, ⬎ 5 years but foreign-born, and U.S.-born subjects. This ordinal variable was significantly associated with being found, by the chi-square test for trend (p ⫽ 0.011) (Table 1). A Table 1 Characteristics of 254 Study Subjects Sought for Follow-up Interview after Release from Jail, by Whether or Not They Were Found Characteristic Gender Male Female Ethnicity Latino Black Other, mixed Asian White Native American Birth country United States Mexico Honduras El Salvador Other Latin America Asia/Pacific Islands Other Preferred language English Spanish Time in United States 0–5 years since immigration ⬎5 years since immigration Born in United States Partnership status Never married, separated, divorced, or widowed Married/partnered Housing before jail Ever homeless in last month Never homeless in last month Employed before jail Employed in last 3 months Unemployed in last 3 months Drugs or alcohol Perceived problem No perceived problem Total a
n
Found (n)
Found (%)
p-value
228 26
167 21
73.2 80.8
ns
147 54 26 14 11 2
97 47 23 9 10 2
66.0 87.0 88.5 64.3 90.9 100.0
0.001a
82 66 37 18 21 17 10
70 38 24 16 16 12 9
85.4 57.6 64.9 88.9 76.2 70.6 90.0
0.001a
109 145
94 94
86.2 64.8
⬍ 0.001
76 84 84
50 60 72
65.8 71.4 85.7
0.011b
167 87
128 60
76.7 69.0
ns
76 175
57 128
75.0 73.1
ns
144 110
109 79
75.7 71.8
ns
136 118 254
99 89 188
72.8 75.4 74.0
ns
Chi-square test of the value in this row as compared to all other values. Chi-square test for trend.
b
244
E. Menendez et al.
stepwise logistic regression analysis indicated that language (preferring English versus Spanish for baseline interview in jail) was the only variable that predicted being found (relative risk: 3.2, 95% confidence interval: 1.5–6.7, p ⫽ 0.002), while controlling for covariates and the clinical trial study group. Time to Locate Subjects and Effective Strategies In the subset of 188 persons who were located, the average time to location was 61.5 days (median ⫽ 36 days, range: 1–180). Of the 21 women found, on average the time to location was 45 days, as compared to 64 days for the 167 men who were located, although this difference was not statistically significant. This sample experienced a high rate of recidivism: overall, nearly one quarter of the subjects (24%) were found back in jail. Phone and personal contacts (family and friends) were the most successful strategies for 53% of the sample. Effective strategies differed among subgroups. Because we found that preferred language, ethnicity, and time in the United States were associated with being found, we described the strategies that were successful, stratified by these variables (Table 2). Data on successful strategies were missing for one subject. Seeking persons in shelters and drug and alcohol programs were successful strategies among non-Latinos and English-speaking subjects. Outreach efforts were successful, in particular among Latinos, Spanish-speaking subjects, and new immigrants. These efforts focused on sections of the city where ethnic groups spent time, including popular restaurants and community gathering places. DISCUSSION In a population of study subjects released from jail, we located 74.0% based on information gathered at baseline. Harlow reported successful tracking to
Table 2 Percent of Persons Found Using Different Search Strategies, by Characteristics Identified as Predictors of Being Found Predictor Language English (n ⫽ 93) Spanish (n ⫽ 94) Time in United States 0–5 years (n ⫽ 50) ⬎ 5 years (n ⫽ 60) Born in United States (n ⫽ 71) Ethnicity Latino (n ⫽ 97) Black (n ⫽ 47) Other, mixed (n ⫽ 22) White (n ⫽ 10) Asian, Native American (n ⫽ 11) Total (n ⫽ 187)
Jail
Phone/Letter
Contacts
Programs
Outreach
26 22
32 34
15 23
18 7
9 13
24 17 30
34 43 24
20 18 20
8 8 20
14 13 7
24 30 18 30 9 24
34 30 32 20 54 33
21 15 23 30 9 20
9 18 27 20 0 12
12 6 9 0 27 11
Contacts included individuals named by study subject. Programs included shelters, drug and alcohol programs, TB clinic, and other programs that subjects provided permission to contact. Outreach included visits to community settings, including popular restaurants and community gathering places.
Locating Study Subjects
245
completion of TB preventive therapy in 63% [20], and our rates of locating subjects were similar to those found by Ludwig et al. [21]. Results from this study focus on locating a specific group that has several differences from those studied by others. Released inmates are a group with special characteristics that may affect outcome. First, because this sample came from a study of TB in a west coast city, it included a disproportionate number of new immigrants, in particular from Latin America, and language was an important predictor of success in locating subjects after release. In addition to cultural differences, they may have had immigration issues that affected their willingness to be found for follow-up. Predictors of locating subjects were not in complete agreement with other research. Poor social functioning, including unpartnered or unmarried status, predicted loss to follow-up in a sample of 85 alcoholics [8]. Although not statistically significant, our findings indicated that persons who were married or partnered at baseline interview were less likely to be located after release from jail. In several instances, the subject named a partner as the main or only contact, and when approached, this contact person was no longer partnered with the former inmate. The fact of being in jail may have prompted the dissolution of the partnership, or, in the extreme, the reason for the person being in jail may have been related to the relationship. Senturia et al. found that the inability to provide several alternative contacts was predictive of retention problems [22]. This finding, combined with our results, leads us to believe that probing for as many contacts as possible is an important strategy, especially if the inmate provides only one partner as a contact. Our work also differed from that of MacKenzie et al. in the area of homelessness. In their study of alcoholics, residential instability in the follow-up period predicted loss to follow-up [8]. In our study, persons who reported homelessness before jail were more likely to be found, although the finding did not reach statistical significance. In this case as well, the fact of the jail stay may have influenced the ease with which persons were found after release. Persons with a history of homelessness were able to name shelters where they occasionally stayed, and these were stable locations as compared to persons who had one or two contact individuals and private residences that may have become disrupted as a result of their jail stay. Among effective strategies, jail itself has been identified as an important setting for locating subjects, in our study and in studies of persons with a history of time in jail. Whelan et al. found that among high-risk drug users in Oregon, 19% were found while they were incarcerated [23], which is similar to our rate of 24% who were found in jail. Ludwig et al. found that the most productive strategies in the community were mail and telephone contacts with relatives and friends [21]. In our sample of released inmates, known to have high rates of substance abuse and mental health problems, programs for drug and alcohol problems were successful places to search for subjects. And finally, in this predominantly Latino sample, our successful outreach efforts included questions about sections of the city where particular subgroups congregated for social or employment reasons. Sending staff to these areas was a successful strategy in particular for those who were new immigrants from Latin American countries. Staff preparation and cultural competence have been cited repeatedly in the
246
E. Menendez et al.
literature and were important in this study. Marmor et al. examined follow-up in the Framingham Children’s Study and found that attitudes of staff, feedback to subjects, and the staff’s handling of questions and problems were important predictors of continued participation [1]. They concluded that quality of the communication with study subjects and the subjects’ perceived importance of the research were key factors in minimizing loss to follow-up. In a study of drug abusers in St. Louis, Cottler et al. identified persistence, creative teamwork, and a comprehensive searching strategy as critical factors in successful follow-up [24]. Krieger et al. found that in a clinical trial in which enhanced searching and follow-up were parts of the intervention, community health workers of the same ethnicity and from the same communities may have accounted for the effectiveness of outreach [25]. Desland and Batey located 76% of heroin users for follow-up interviews, citing collection of good and reliable information, ethical practices in informing subjects of searching methods, confidentiality of the methods, and using creativity and flexibility in searching for subjects as reasons for their success [3]. In our study, research personnel were bilingual and bicultural, search protocols were consistent and monitored, and the study team was trained, monitored, and encouraged and maintained enthusiasm throughout the study. We believe these are critically important aspects of a study protocol, both to obtain sufficient contact information and to approach contacts appropriately and successfully in the community. Locating study subjects in longitudinal research is critical to achieve sufficient sample size for analysis and to make conclusions that can be considered valid. The population of released inmates shares characteristics, such as unstable housing, drug and alcohol problems, and unpartnered status, that have been identified as predictors of loss to follow-up. However, in this study these characteristics did not predict loss to follow-up, as incarceration per se seemed to influence the ability to locate study subjects. For this population, therefore, attention should be paid to collection of adequate residential and social contact information. Specific ethnic groups, such as Latinos in this study, differ in the social contacts and strategies that will be most successful. For those who prefer a language other than English, special efforts include employment of bicultural as well as bilingual personnel. Identification of those who are likely to be difficult to find can help project managers allocate resources appropriately. This project was funded by the National Institutes of Health, NINR (R01 NR04456).
REFERENCES 1. Marmor J, Oliveira S, Donahue R, et al. Factors encouraging cohort maintenance in a longitudinal study. J Clin Epidemiol 1991;44:531–535. 2. Freedman D, Thornton A, Camburn D. Maintaining response rates in longitudinal studies. Soc Methods Res 1980;9:87–98. 3. Desland M, Batey R. High retention rates within a prospective study of heroin users. Br J Addict 1991;86:859–865. 4. Goldstein PJ, Abbott W, Paige W, Sobel I, Soto F. Tracking procedures in follow-up studies of drug abusers. Am J Drug Alcohol Abuse 1977;4:21–30. 5. Bale R, Cabrera S, Brown J. Follow-up evaluation of drug abuse treatment. Am J Drug Alcohol Abuse 1977;4:233–249.
Locating Study Subjects
247
6. Moos R, Bliss F. Difficulty of follow-up and outcome of alcoholism treatment. J Stud Alcohol 1978;39:473–490. 7. Klonoff-Cohen H. Tracking strategies involving fourteen sources for locating a transient study sample: Parents of sudden infant death syndrome infants and control infants. Am J Epidemiol 1996;144:98–101. 8. MacKenzie A, Funderburk FR, Allen RP, Stefan R. The characteristics of alcoholics frequently lost to follow up. J Stud Alcohol 1987;48:119–123. 9. Pollio DE, Thompson SJ, North CS. Agency-based tracking of difficult-to-follow populations: Runaway and homeless youth programs in St. Louis, Missouri. Community Ment Health J 2000;36:247–258. 10. Beck AJ. Prison and jail inmates at midyear 1999. In: Bureau of Justice Statistics Bulletin. Washington, D.C.: U.S. Dept of Justice; April, 2000. NIJ 181643. 11. Wright J, Lampton A, Devine JA. Tracking non-traditional populations in longitudinal studies. Evaluation and Program Planning 1995;18:267–277. 12. Given B, Keilman L, Collins C, Given W. Strategies to minimize attrition in longitudinal studies. Nurs Res 1990;39:184–186. 13. Hough R, Tarke H, Renker V, Shields P, Glatstein J. Recruitment and retention of homeless mentally ill participants in research. J Consult Clin Psychol 1996;64:881–891. 14. McKenzie M, Peterson Tulsky J, Long HL, Chesney M, Moss A. Tracking and follow-up of marginalized populations: A review. J Health Care Poor Underserved 1999;10:409–429. 15. Areán PA, Gallagher-Thompson D. Issues and recommendations for the recruitment and retention of older ethnic minority adults into clinical research. J Consult Clin Psychol 1996;64:875–880. 16. Craig R. Locating drug addicts who have dropped out of treatment. Hosp Community Psychiatry 1979;30:402–404. 17. Eckland BK. Retrieving mobile cases in longitudinal surveys. Public Opinion Quart 1968;32:51–64. 18. Tulsky JP, White MC, Dawson C, et al. Screening for tuberculosis in jail and clinic follow-up after release. Am J Public Health 1998;88:223–226. 19. White MC, Tulsky JP, Reilly P, et al. A clinical trial of a financial incentive to go to the tuberculosis clinic for isoniazid after release from jail. Int J Tuberc Lung Dis 1998;2:506–512. 20. Harlow T. TB net tracking network provides continuity of care for mobile TB patients. Am J Public Health 1999;89:1581–1582. 21. Ludwig AM, Levine J, Stark LH. LSD and Alcoholism: A Clinical Study of Treatment Efficacy. Springfield, Illinois: Thomas, Inc.; 1970. 22. Senturia Y, Mortimer K, Baker D, Joseph C, Wedner J. Successful techniques for retention of study participants in an inner city population. Control Clin Trials 1998;19:544–554. 23. Whelan V, McBride D, Colby R. Public health department tracking of high-risk drug users. Public Health Rep 1993;108:643–645. 24. Cottler L, Compton W, Ben-Abdallah A, Horne M, Claverie D. Achieving 96.6 percent follow-up rate in a longitudinal study of drug abusers. Drug Alcohol Depend 1996;41:209–217. 25. Krieger J, Collier C, Song L, Martin D. Linking community based blood pressure measurement to clinical care: A randomized controlled trial of outreach and tracking by community health workers. Am J Public Health 1999;89:856–861.