INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Socially desirable responding is the tendency of an individual to answer questions in a manner that conforms to social norms or to give a favorable impression of oneself (1). Several studies have found higher rates of socially undesirable behavior reported among respondents using computer-assisted questionnaires when compared with those using interviewer-assisted questionnaires (2). In addition, computer-assisted questionnaire techniques have been cited as a means to reduce processing costs and increase interview efficiency (4) while offering an appropriate interview tool for illiterate or multiply ethnic populations (5).

Psychometric properties of computer-assisted questionnaire responses have been deemed comparable to those of interviewer-assisted responses among injection drug users (IDUs) (6). In addition, out-of-treatment drug users reported adequate skills to successfully complete a computer-assisted questionnaire, and had no reservations regarding confidentiality provided by such methods (7). In randomized studies, computer-assisted techniques have shown higher rates of risky drug use behaviors reported by men who have sex with men (8), and higher rates of risky human immunodeficiency virus (HIV)-associated behaviors reported by IDUs (9) compared with interviewer-assisted techniques. These results support the hypothesis that computer-assisted techniques resulted in more accurate reporting. Randomized studies that have not found higher rates of risky behavior among computer-assisted respondents still maintain that computer-assisted methods are reliable (10), may enhance the quality of behavioral assessment, and offer an acceptable means by which to collect self-reports of HIV risk behavior (11).

To date, no studies have compared computer-assisted with interviewer-assisted questionnaires with regard to risk factors for tuberculosis. Tuberculosis risk factors are largely related to socioeconomic status (12), homelessness (13), smoking (14), injection drug use (12, 15), and HIV infection (16, 17). It has been suggested that "shotgunning," the practice in which an individual inhales smoke and exhales into the mouth of a second person, offers another potential route for efficient transmission of respiratory pathogens (18). Injection drug use and homelessness have also been associated with HIV, suggesting an overlap between HIV and tuberculosis (19). Additionally, although cigarette smoking has not been associated with the progression of HIV infection (20), it has been associated with HIV-related opportunistic infections (21), as well as with suppressing lung immune response (22, 23). Crack smoking has been cited as another risk factor for HIV (24) and has been linked to outbreaks of tuberculosis through the use of crack houses (25).

Because tuberculosis and HIV are closely linked, improved HIV research methods would likely improve the quality of tuberculosis research. Such improvements could lead to more accurate characterization of tuberculosis risk factors, resulting in more relevant and better designed prevention programs. The current study compared tuberculosis and HIV risk factors between computer-assisted and interviewer-assisted questionnaires among individuals requesting tuberculosis screening at sites of the Baltimore needle exchange program (NEP). The hypothesis of this study was that respondents completing computer-assisted questionnaires would report higher rates of tuberculosis and HIV risk behaviors than respondents completing interviewer-assisted questionnaires.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Between June 1998 and May 1999, tuberculosis screening and directly observed therapy were offered to participants of the Baltimore NEP (denoted as the TB/NEP study). Tuberculosis education, screening, skin test assessment, and directly observed therapy were available to all NEP participants who had no history of tuberculosis therapy and who had not been screened for tuberculosis during the prior 6 mo. Restricting testing services to individuals who had not been tested during the prior 6 mo was an attempt to minimize excessive testing by persons wishing to repeatedly receive monetary incentive. No active recruitment took place for the TB/NEP study. Rather, signs were posted and NEP staff described the service to participants who either chose to participate or not to participate. At the time of tuberculin testing, TB/NEP participants underwent a consent procedure, were administered a TB/NEP-specific interview, and asked to return in 72 h for tuberculin skin test reading. Upon returning for their skin test reading, participants received five dollars (United States).

The TB/NEP interview consisted of 80 questions regarding sociodemographic factors, current injection drug use, tuberculosis history, current tuberculosis risk factors, tuberculosis knowledge, HIV risk factors, HIV status, physical functioning, and health care utilization during the prior year. Before implementation, the TB/NEP questionnaire was pilot tested to ensure optimal questionnaire design, appropriate language, and ease of computer use. The same questionnaire was administered to all participants; however, the first 190 (67%) completed an interviewer-assisted questionnaire, whereas the remaining 92 (33%) underwent a computer-assisted questionnaire (i.e., listened to questions through headphones and responded using a computer keyboard). Study staff gave the computer-assisted respondents a 5-min introduction to the computer, the keys that would be used, and the types of questions that would be asked. In addition, respondents were told that they could call study staff into the room for technical assistance at any time. This research protocol was approved by the Committee on Human Research at the Johns Hopkins School of Public Health.

Interview questions were identical in both the computer-assisted and interviewer-assisted questionnaires. Response patterns were categorical and continuous. Computer-assisted respondents answered questions by pressing a numeric computer key corresponding to the code of the desired response (e.g., "if you have ever smoked heroin, press the number one key now"). Dichotomous variables were created from continuous variables by categorizing at the median of the continuous variable. Odds ratios (OR) and confidence intervals (CI) were calculated to determine the magnitude of interview technique effect, as well as the amount of variability in each estimate. Inferences were based on simultaneous adjustment for independent variables using multiple logistic regression in which all bivariate characteristics with CI that did not include one were entered into a multivariate model. Independent variables were deleted from the logistic regression model if they did not contribute to the overall fit of the model. Goodness of fit was determined by log likelihood ratios. Interaction terms were also tested. Models were created first within groups of similar characteristics (i.e., sociodemographics, smoking characteristics, and drug use characteristics), and then for all significant characteristics. Separate models were created to show the effects of simultaneous adjustment within and between characteristic groups. Among participants who were screened twice during the 11-mo study period, only the first interview was included for analysis. This was done to preserve independence of the data.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Among 282 participants of the TB/NEP program, 190 completed interviewer-assisted questionnaires and 92 completed computer-assisted questionnaires. The average time taken to complete an interviewer-assisted questionnaire was 30 min; the average time taken to complete a computer-assisted interview was 45 min. Although 10 individuals were tested and interviewed twice during the 11-mo program, only first interviews were included in the current analyses. The median age of respondents was 40 yr (interquartile range = 36 to 46), 92% were African-American, 41% were female, 39% graduated high school, and 8% were employed at the time of registration. These characteristics did not differ significantly between computer-assisted and interviewer-assisted questionnaire groups. Likewise, no significant differences between interview groups existed with respect to self-reported health status, HIV status, drug treatment participation, daily alcohol intake, and reporting use of homeless shelters or soup kitchens during the past year. (see Table E1 in the online data supplement).

The vast majority of this study population had smoked cigarettes, marijuana, or crack; however, only 21% had smoked heroin (Table 1). Respondents using computer-assisted questionnaires were more likely to report smoking marijuana (OR = 7.66), crack (OR = 2.86), and heroin (OR = 3.36) (Table 1). Although there were no differences with respect to reports of sharing cigarettes, respondents completing computer-assisted questionnaires were four times more likely to report sharing cocaine smoking equipment (OR = 4.0) and heroin smoking equipment (OR = 3.95) than those completing interviewer- assisted questionnaires (Table 1). Reports of shotgunning were also proportionately higher among computer-assisted questionnaire respondents (OR = 6.11). After simultaneously controlling for all significant smoking characteristics, variables that differed by interview technique were smoked marijuana, smoked crack, smoked heroin, shared cocaine smoking equipment, and shotgunned (Table 1).

The median duration of injection drug use was 20 yr for this population (interquartile range = 10 to 27), median number of NEP visits during the 11-mo study period was 15 (interquartile range = 5 to 38), and 82% of the population injected daily. These characteristics did not significantly differ by interview method (Table 2). However, computer-assisted questionnaire respondents were more than twice as likely to report staying in the same place and getting high with the same people for several days at a time ("binging") (OR = 2.24), visiting a shooting gallery (OR = 3.10), and visiting a crack house (OR = 5.80) during the past year, than were interviewer-assisted questionnaire respondents (Table 2). After simultaneously controlling for these three characteristics, only visiting a crack house differed by interview technique (Table 2).

Finally, a multivariate model was constructed in which all significant variables were included. After simultaneously controlling for smoked marijuana, smoked crack, smoked heroin, shared cocaine smoking equipment, shotgunned, and visited a crack house, computer-assisted questionnaire respondents were more than twice as likely to report visiting a crack house (adjusted OR = 2.69), and more than 5 times as likely to report shotgunning (adjusted OR = 5.18) than were their interviewer-assisted questionnaire counterparts (see Table E2 in the online data supplement).

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

This was the first study to compare interviewer-assisted with computer-assisted questionnaire interview techniques with regard to tuberculosis risk factors. Comparing interviewer-assisted questionnaires with computer-assisted questionnaires, we found no differences with regard to sociodemographic or access to healthcare services characteristics, but observed higher reports of visiting a crack house and "shotgunning" among respondents who completed computer-assisted questionnaires compared with those who completed interviewer-assisted questionnaires. These data show that self-reports of drug use behaviors associated with tuberculosis and HIV infection are consistently reported at higher levels when computer-assisted interviewing techniques are employed.

In other studies, computer-assisted questionnaires have also shown higher levels of reporting for risky behaviors (8, 9). However, few studies that validate these types of results to an external source of information currently exist. Several points support the assertion that the higher reports of sensitive behaviors we observed on computer-assisted questionnaires relative to interviewer-assisted questionnaires did in fact more accurately represent the true practices of this population. First, nonsensitive variables (i.e., sociodemographic factors and health/ access) did not differ by interview technique, whereas higher risk variables were reported more frequently among computer-assisted questionnaire respondents. Second, differences between interview techniques were consistent in direction, with computer-assisted questionnaire respondents reporting more frequent risky behavior. Third, our data are consistent with similar studies (8, 9). The patterns of response found in previous studies and the current study are consistent with the inference that data from computer-assisted questionnaires are less influenced by socially desirable responding than data from interviewer-assisted questionnaires. Additionally, no peer-reviewed studies currently indicate a reason or situation in which individuals would overreport undesirable behavior. Assuming that computer-assisted questionnaire reports more accurately represent the practices of the population, a reliance on interviewer-assisted questionnaires would lead to underestimating crack house visits and shotgunning by more than 30%. Such differences could have implications for interpreting previous research findings regarding tuberculosis and HIV risk factors.

Possible limitations to the generalizability of this study include a small sample size, the fact that all respondents chose to use health care services at sites of needle exchange, and the fact that study respondents were not randomized to mode of interview. It is possible that the first 190 interviewer-assisted questionnaire respondents sought health services more aggressively than the remaining 92 computer-assisted respondents, thus biasing results. However, because physical functioning and drug treatment participation did not significantly differ by interview technique, this possibility appears to be unlikely. In addition, there is no reason to suspect that visitors of crack houses or those individuals who shotgunned would have been more likely to use tuberculosis screening sooner than those who did not practice these risky behaviors. All interviewing took place within an 11-mo study period, thus temporal affects should have been minimal. A major strength of the study is that respondents were current IDUs, recruited from sites of NEP, making the results applicable to active injectors outside of institutional settings.

Valid measurement of socially sensitive behaviors is necessary to accurately assess risk factors for HIV and tuberculosis. Coupled with earlier reports regarding HIV risk behaviors (8, 9) the current study supports the use of computer-assisted questionnaires as a method to increase reporting of sensitive behavioral data for tuberculosis and HIV risk factors.