How good is fine-grained Timeline Follow-back data? Comparing 30-day TLFB and repeated 7-day TLFB alcohol consumption reports on the person and daily level

Abstract

Objective

This study examined the correspondence of two types of Timeline Follow-back (TLFB) methods, a web-based self-administered, repeated 7-day TLFB and an interviewer-administered 30-day TLFB of alcohol consumption.

Method

Participants were first- and second-year college students (n = 323, 58.5% female). Day-to-day correspondence of drinking reports and correspondence of person-level indicators of drinking were assessed.

Results

Results indicated that correspondence between the TLFB-30 and TLFB-7 reports was generally good for summary indicators of drinking, but TLFB-7 data indicated a statistically significantly higher number of total drinks consumed, a higher number of days drinking 4+/5+ drinks per day, and a lower number of abstinent days than TLFB-30. Similarly, day-to-day comparison of drinking reports showed that drinking days were more frequently reported using the TLFB-7, a trend which was more pronounced for distal weekdays than recent weekdays. Correlations between TLFB-7 and TLFB-30 reports of drinks per drinking day were also lower for distal compared to recent weekdays (r = 0.61 vs. r = 0.76). Using a Poisson regression model, a linearly increasing trend in the absolute value of the difference between TLFB-7 and TLFB-30 drinking reports per day as length of recall increases was found (b = 0.013, z = 4.43, with p < 0.001).

Conclusions

Our results indicate that participants reported more drinking on the repeated TLFB-7 than on the standard TLFB-30. Furthermore, the result of daily level analyses showed that discrepancies between the methods increased as the length of recall increased. These findings suggest that TLFB assessments covering longer intervals may have reduced accuracy on a fine-grained scale.

Introduction

The Timeline Follow-back (TLFB) interview is one of the most widely used measures of alcohol consumption. Originally developed as an interviewer-administered paper-and-pencil measure, the TLFB collects retrospective estimates of daily drinking over a specified period of up to 12 months. Memory aids are provided to prompt recall, such as a calendar format and the provision of important dates. The TLFB has been used for over 30 years in clinical and nonclinical research studies (Sobell, & Sobell, 2008), and its cross-cultural validity has been demonstrated in several countries (i.e., Australia, Canada, Mexico and Spain; Sobell et al., 2001). Recent research further documents and strengthens the evidence supporting this measurement's construct validity compared with other retrospective drinking assessments, such as the Quick Drinking Screen in samples of outpatient alcohol abusers (Roy et al., 2008) and non-clinical individuals with alcohol problems (Sobell et al., 2003).

The cost of using a skilled interviewer for the TLFB is considerable (Maisto, Conigliaro, Gordon, McGinnis, & Justice, 2008), which has prompted the development of alternative administration methods. To date, several variations of the original interviewer-administered paper-and-pencil assessment exist, where assessment costs are reduced through the use of technology to either eliminate the need for in-person interviews through the use of phone interviews or by replacing the interviewer altogether with automated instructions via a computer program or voice-interactive system. Few studies, however, have documented the impact of different administration methods on the reporting of alcohol consumption.

Two studies have examined the effect of self- and other-administration methods. The first study (Sobell, Brown, Leo, & Sobell, 1996) examined the effect of different self-administration methods (i.e., paper vs. computer) and other-administered methods (i.e., in-person vs. via phone) in two samples. The first sample (n = 63 problem drinkers entering treatment) completed a paper-and-pencil version of the TLFB, describing their drinking during the 90 days pre-treatment. Approximately 3 weeks later, they used a computer program with the identical instructions to complete another TLFB assessment for the same days. The second sample (n = 40 problem drinkers who had agreed to participate in a treatment study) completed a 30-day interviewer-guided TLFB via phone one month after treatment, and an interviewer-guided in-person TLFB assessment assessing drinking behavior over the 12 months after treatment, which included the 30 days assessed in the first TLFB assessment. Results from both samples indicated no differences in drinking between administration methods, where total number of drinks, days abstinent, days drinking 1–4 drinks, days drinking 5–9 drinks, days drinking > 10 drinks, and drinks per drinking day were compared. The authors concluded that reliable drinking data can be obtained using these different methods for interviewer-administration and self-administration of the TLFB.

A more recent experimental study (n = 73 participants of the Veterans Aging Cohort Study; (VACS: Justice et al., 2006)) compared paper and pencil self-administration to interviewer-administration conducted via phone (Maisto et al., 2008). Participants completed two assessments of 30-day TLFB, completed no more than one week apart. They were randomized into four groups of assessment order (i.e., self–self, self–telephone, telephone–self, and telephone–telephone). This study as well reported has no differences between administration methods, where the indices of interest were the median number of total drinks and percentage of hazardous drinkers.

Notably, research on the impact of different administration methods on TLFB data has focused on person-level indices of drinking that summarize drinking into static variables, such as the median number of drinks across a specified timeframe. Much less is known about the day-to-day accuracy of TLFB methods, such as the number of drinks consumed on particular weekdays, for example. Accuracy on such a fine-grained scale is critical, however, if data are to be analyzed on that level. If the goal of a study lies in comparing two groups in their volume of alcohol consumption, such level of detail is not necessary. If, however, the goal is to understand and capture processes and mechanisms of change, the fine-grained details of alcohol consumption matter. A particular strength of the TLFB is its ability to capture sporadic and unpatterned drinking (Sobell et al., 2003), where both volume and timing of drinking are relevant. Indeed, this feature of the TLFB is particularly relevant in light of the growing evidence that shows that the pattern of drinking is an important determinant of the impact of alcohol use on health, independent of the volume of alcohol consumed (Bobak et al., 2004, Rehm et al., 2001, Room et al., 2003, Russell et al., 2004, Tolstrup et al., 2004). As such, the accuracy of different administration methods of the TLFB on a day-to-day level becomes relevant.

Only a handful of studies have examined the accuracy of the TLFB on a day-to-day basis, where TLFB methods have been compared to real-time and daily recall methods. Carney, Tennen, Affleck, Del Boca, and Kranzler (1998) compared 28-day TLFB, ecological momentary assessment (EMA) and daily diary reports of alcohol consumption in two samples. Results indicated that compared to daily diary (n = 20 problem drinkers) and EMA (n = 48 moderate drinkers), 28-day TLFB data identified fewer drinking days, fewer drinks per drinking day and fewer total drinks, yet that these differences were modest. Of greater concern were the large individual differences in day-to-day correspondence, which the authors noted, suggested that the 28-day TLFB may be less useful for detecting patterns of consumption. Similar disparities in correspondence between day-to-day patterns of drinking were found when TLFB methods were compared to daily interactive voice mail reports, where correlations between daily self-reports and 28-day TLFB ranged from − 0.22 to 0.96, with an average of 0.51 (Perrine, Mundt, Searles, & Lester, 1995) and from 0.54 to 0.79 when compared to 7-day TLFB (Toll, Cooney, McKee, & O'Malley, 2006). These studies suggest that the TLFB may be limited in the accuracy with which it assesses fine-grained characteristics of alcohol consumption, particularly as the length of recall increases.

Promisingly, in addition to changing the type of administration of the TLFB, technology (e.g., internet surveying) has also made it possible to increase the frequency of TLFB administration and to shorten the length of recall intervals. Such repeated, short-interval TLFB assessments are likely subject to fewer recall biases than longer TLFB assessments. They also constitute a different type of response burden for the participant. Past research has shown that comprehensive assessments, such as the 12 month TLFB, are associated with a much lower completion rate than brief measures of alcohol consumption (22% vs. 51% for a mailed out survey) (Cunningham, Ansara, Wild, Toneatto, & Koski-Jannes, 1999). By comparison, the short and frequent reports of EMA measures of alcohol consumption have often shown compliance rates upward of 90% (e.g., Litt et al., 2000, Todd et al., 2009), though they can also fall below 50% (e.g., Litt, Cooney, & Morse, 1998). A short-term, frequently repeated TLFB is less frequent than EMA but can be similarly brief, and thus might potentially result in better compliance than more comprehensive TLFB assessments. This study focuses on such short-term, frequently repeated TLFB and compares it to the more traditionally used 30-day TLFB.

The purpose of this study was to evaluate the correspondence of a web-based self-administered, repeated 7-day TLFB and an interviewer-administered 30-day TLFB of alcohol consumption in a sample of first- and second-year college students. Of particular interest was the day-to-day correspondence of drinking reports in addition to providing information about the correspondence of person-level indicators of drinking. A focus on day-to-day correspondence addresses the accuracy of TLFB data not just in regards to the volume but also the pattern of drinking. In first- and second-year college students, who are underage drinkers and thus less likely to have established drinking patterns in that they intersperse days of abstinence and excessive drinking (Collins, Kashdan, Koutsky, Morsheimer, & Vetter, 2008), the timing of drinking events is particularly important, rendering a comparison of methods in this population especially salient. We hypothesized that the repeated, short-term TLFB method would result in reports detailing a higher frequency and volume of alcohol consumption, particularly as length of recall increases.