Drug and Alcohol Dependence, 9 (1982) 101 - 110 101

OVER-REPORTS OF RECENT ALCOHOL CONSUMPTION IN A CLINICAL POPULATION: A VALIDITY STUDY

LORRAINE MIDANIK

Alcohol Research Group (formerly Social Research Group), Institutes ofMedical Sciences,

Institute of Epidemiology and Behavioral Medicine, 1816 Scenic Avenue, Berkeley, CA 94709 (U.S.A.)

(Received December 15, 1981)

Summary

Decisions concerning appropriate treatment in alcoholism programs are often based on the self-reports of the clients. However, few programs have incorporated validation procedures (such as breath tests) since it is generally assumed that alcoholics will deny the extent of their drinking and their sub- sequent alcohol problems.

The self-reports of recent alcohol consumption of sixty-five new entrants to an alcoholism treatment program were validated with breath tests. Five estimates of blood alcohol concentrations were derived by varying elimination rates. For all five estimates, over-reporters comprised a substan- tial proportion of the total sample (23% - 57%) and exceeded the percentage of consistent reporters for those people who had a positive breath test. Cor- relations between self-reports and breath tests were not significant which indicates that the amount of alcohol consumed does not necessarily relate to reporting behavior.

It is concluded that despite the widely held notion that alcoholics deny the extent of their drinking, errors in the direction of over-reporting should be taken seriously and examined more closely. On an individual level, over- presentation of a client’s condition may be related to subsequent behavior in a treatment program; on an aggregate level, over-reports may bias the findings of evaluation studies by inflating success rates.

Introduction

In a recent review of validity studies in the alcohol field, Midanik [l] concluded that the accuracy of self-reports of alcohol consumption and alcohol problems varies with the type of behavior being measured, the level of analysis chosen and the criterion by which the behavior is being judged. Unfortunately many behaviors associated with alcohol use, such as drinking patterns over time and specific withdrawal symptoms, are not easily vali- dated by currently used methods. However, clients’ accounts of recent

0376-8716/82/0000-0000/$02.75 @ Elsevier Sequoia/Printed in The Netherlands

102

alcohol consumption, which are often the basis of treatment decisions in alcohol programs, are easily validated by breath or blood tests which deter- mine blood alcohol levels. Yet few programs incorporate such a procedure as a routine method of assessing the condition of new program entrants. This is surprising since there is evidence to suggest that self-reports of alcohol consumption may be under-estimated when clients are drinking and even trained observers cannot identify intoxicated clients with a high degree of accuracy [ 21 .

The purpose of this study was to examine the relationship between self-reports of recent alcohol consumption and blood alcohol concentrations as determined by breath tests in a group of new clients entering an alco- holism treatment program. Data on recent drinking is generally collected in a summarizing manner, e.g. clients are asked the total amount of alcohol consumed over the previous 24 hours. In an attempt to alleviate reporting errors previously found in this type of validity study a very detailed inter- view schedule was designed to elicit information on alcohol consumption within very specific timeframes during the 24-hour period before the interview.

Methods

Subjects The study population consisted of 68 individuals entering a compre-

hensive alcoholism treatment clinic for the first time during the period of June 20 through November 30, 1977. Forty-nine respondents were inter- viewed in the emergency room (ER); nineteen were seen in the walk-in clinic (WI). Table 1 presents the demographic characteristics of the respon- dents by site of entrance into the program. In general, both samples consistec

TABLE 1

Demographic characteristics of study respondents by site of entrance into treatment program

Emergency Room Walk-In Clinic Test of Significance

Age (mean years) 36.2 (N=49) Race (% black) 67.3 (N=49) Sex (% male) 93.9 (N=49) Education (mean years) 9.3 (N=36) Employment (% unemployed) 81.1 (N=30) Marital Status

% married 18.4 (N=38) % separated, widowed, divorced 55.3 (N=38) % never married 26.3 (N=38)

35.8 (N=19) t = 0.13 NS 84.2 (N=19) x2 = 1.19 NS 78.9 (N=19) x2 = 1.89 NS

9.3 (N=18) t = 0.04 NS 66.7 (N=18) x2 = 0.19 NS

26.3 (N=19) x2 = 5.86 NS 68.4 (N=19)

5.3 (N=19)

NS = not significant.

103

of fairly young (average age approximately 36 years), predominantly black males who tended to be unmarried and unemployed with an average of nine years of education. Since the ER entrants and the WI entrants did not differ significantly in their demographic characteristics, further analysis will com- bine both samples.

Procedures After routine clinical procedures, each eligible respondent was asked

to participate in the research study. Counselors in the clinic obtained signa- tures on the first consent form which explained the purpose of the study and asked for their participation in an interview about their recent drinking behavior. After the questionnaire was completed by either the researcher or a counselor, a second consent form was read to the respondent which explained the breath test procedures. After it was signed, a breath test was administered. Respondents were unaware during the initial interview that they would be asked to give a breath sample. Of the 68 people interviewed, 3 individuals were unable to report specific amounts of alcohol consumed and were therefore eliminated from the analysis.

As the study questionnaire was designed to elicit very specific infor- mation on alcohol consumption during the previous 24-hour period, space was provided to record drinking amounts hourly for the most recent 8-hour period and in 4-hour segments for the remaining 16 hours. It was hoped that these timeframes would enable the respondent to remember his/her recent drinking more accurately. A more detailed description of the procedures can be found elsewhere [3] .

Breath test procedures The Alcolmeter, manufactured by Intoximeters, was the breath

screening device used for the study. This portable unit is about the size of a pocket calculator and weighs approximately 5 ounces (140 g). The Alcol- meter contains a fuel cell sensor backed by a spring loaded syringe-sampling valve which, when released, draws one cubic centimeter sample from the breath into its fuel cell for analysis. The fuel cell is a chamber partitioned by a plastic membrane which is coated with a thin layer of an active chemi- cal. When a breath sample is drawn into the cell, all alcohol is absorbed on the membrane and converted to acetic acid within a short period. The re- sulting electric current, proportionate to the amount of alcohol on the membrane, is displayed on the drum or meter in front of the instrument. The maximum blood alcohol concentration (BAC) reading takes 15 - 60 seconds to develop. The highest reading on the Alcolmeter is 0.30 [4] .

In order to ensure a valid and reliable reading from the Alcolmeters, the instruments were calibrated weekly using a Mark 11A Simulator manufac- tured by Smith & Wesson. In addition, the Alcolmeters were compared to the Breathalyzers used by the Maryland State Police Department in a con- trolled drinking experiment in April, 1978. The average absolute difference between the Alcolmeters and the Breathalyzer for eleven separate trials was

104

0.013%. In eight of the eleven trials, the Alcolmeters registered a slightly lower reading than the Breathalyzer.

Estimation of blood alcohol concentrations Self-reported information on recent alcohol consumption obtained from

the study questionnaire was used to calculate the Estimated Blood Alcohol Concentration (EBAC). Figure 1 presents the specific formula used for this analysis. To derive the EBAC for each client first the number of ounces of absolute alcohol consumed is calculated (number of ounces of alcoholic drink multiplied by the percent of alcohol, e.g. beer 4%, table wine 12%, fortified wine 20%, liquor 40%) and totaled for each beverage type. This total is then multiplied by the specific gravity of alcohol (0.8) which is defined as the ratio between the density of alcohol and the density of dis- tilled water at 4 “C. The denominator of the formula is the product of the respondent’s weight in ounces and Widmark’s R. Widmark’s R is defined as the distribution ratio of alcohol in the body as a whole to alcohol in the blood: values are 0.68 for men and 0.55 for women [ 51. The result of di- viding the numerator by the denominator in this formula, when expressed as a percent, yields maximum estimates of blood alcohol concentrations since no elimination rates have been considered [ 61.

(#Of0 zs. EBAC =

of Alcohol Consumed) x % Alcohol X (Specific Gravitya) Total

Weightb x Widmark’s RC - Elimination

Rated

t6.8 for ethyl alcohol m ounces

‘0.68 for men; 0.55 for women d(BAC reduction per hour x (#ofh ours since alcohol

ranging from 0.010 -- 0.030%) consumed in last 24 hours)

Fig. 1. Formula for Estimated Blood Alcohol Concentrations (EBAC).

The last element of the formula is perhaps the most controversial in the literature, i.e. elimination rates. Alcohol is eliminated from the body primar- ily by metabolism and secondarily through excretion. Since only a small amount of alcohol is eliminated by excretion in breath, urine, tears, saliva, etc., the issue is how to estimate metabolic elimination accurately. In specific case studies there is evidence to suggest that the elimination rate increases as the blood alcohol concentration (BAC) increases [ 71. Higher elimination rates appear to apply to extremely high concentration levels (0.30% - 0.40%). Given that this relationship. is valid, these higher rates are not applicable to this study since the study sample did not exceed a BAC of 0.27%. The issue of using average elimination rates is slightly more complex. While the average rate of BAC reduction is reported at 0.015% to 0.018% per hour for general populations, this range can vary considerably. Therefore the data will be presented using EBACs based on five estimates of elimination rates (ranging from 0.010% to 0.030% BAC reduction per hour). Comparisons can then be

105

made of discrepancies which occur between self-reports and actual BACs when the estimating formula is varied. These varying rates will then be multi- plied by the number of hours during which alcohol was consumed to obtain the total elimination rate for each respondent during the 24-hour period prior to the initial contact.

The formula just presented assumes a zero BAC at the start of the 24-hour period. In cases where drinking was spaced so that there were sev- eral drinking episodes within the 24-hour period allowing enough time to reach a zero BAC between episodes, EBACs were calculated on the most recent drinking episode prior to which a zero BAC can be determined.

Results

Table 2 presents the distributions of absolute alcohol consumed as measured by self-reports and the breath tests (without adjustment for elimi- nation rates) for the total sample and for clients whose BACs were greater than zero. For both samples the means, medians, standard deviations and ranges differ greatly between the BAC reading and the self-report. For the total sample, the mean ounces of absolute alcohol derived from the self- reports is almost eight times greater than that obtained from the breath test (10.43 oz. as compared to 1.39 oz.). Since the mean is affected by extreme cases and the dispersion is quite large as indicated by the standard deviations for both groups, the median is more appropriate as a measure of central ten- dency for this sample. Here again it is apparent that the distributions are very different with self-reported alcohol consumption reflecting a much greater amount (9.04 oz. as compared to 0.00 oz.). When those with zero blood alcohol concentrations (as measured by the breath test) are eliminated from the analysis, the relationship between the two measures remains the same with self-report yielding a much higher mean and median than the breath test (Mean16.72 oz. versus 3.35 oz.; Median-13.12 oz. versus 3.04 oz.).

TABLE 2

Means, medians, standard deviations and ranges of total ounces of absolute alcohol consumed as calculated from self-reports and breath tests

Self-reports Breath tests - Total Sample Clients with Total sample Clients with (N=65) positive BAGS (N=65) positive BACs

(N=27) (N=27)

Mean ounces 10.43 16.72 1.39 3.35 Median ounces 9.04 13.12 0.00 3.04 Standard Dev. 11.96 12.93 2.10 2.03 Range 0.0 - 55.28 0.0 - 55.28 0.0 - 6.82 0.19 - 6.82

106

Based on the data just presented we can look even further at the rela- tionship between the BAC derived from the Alcolmeter and the estimates of BAC based on the formula presented in Fig. 1 using the five different elimi- nation rates. For the rest of the analysis EBAC1.a will represent estimates using the following elimination rates: EBAC, = 0.010% reduction per hour, EBACs = 0.015% reduction per hour, EBACs = 0.020% reduction per hour, EBACI = 0.025% reduction per hour and EBA& = 0.030% reduction per hour. Table 3 presents the means, medians, standard deviations and ranges for BACs (as obtained from breath tests) and each EBAC (from self-reports) for the total sample (N=65) and for those people who had a positive BAC (N=27). Despite the elimination rate used, the means and medians of all five EBACs for the total sample are greater than the mean BAC. This pattern is also evident when only those respondents with positive BACs are exam- ined separately. For this group the means of both the BAC and EBACs are inflated but the mean ounces of absolute alcohol consumed as measured by self-reports (EBAC) is still much larger than the breath test results.

TABLE 3

Means, medians, standard deviations and ranges for BACs (from breath tests) and Esti- mated BACs (from self-reports) for total sample and for clients with positive BACs

Total sample (N=65)

Clients with positive BACs (N=27)

mean sd median range mean sd median range BAC 0.060 0.090 0.000 0 - 0.27 0.145 0.086 0.150 0.01 - 0.27 EBAC, * 0.342 0.450 0.180 0 - 1.91 0.562 0.529 0.460 0 - 1.91 EBACz 0.278 0.415 0.070 0 - 1.79 0.471 0.504 0.340 0 - 1.79 EBAC, 0.223 0.380 0.001 0 - 1.67 0.390 0.473 0.220 0 - 1.67 EBAC4 0.176 0.344 0.000 0 - 1.55 0.314 0.441 0.100 0 - 1.55 EBAC, 0.136 0.309 0.000 0 - 1.43 0.247 0.408 0.000 0 - 1.43

*EBAC1 is based on 0.010% BAC reduction per hour. EBACz is based on 0.015% BAC reduction per hour. EBACa is based on 0.020% BAC reduction per hour. EBAC, is based on 0.‘025% BAC reduction per hour. EBACs is based on 0.030% BAC reduction per hour.

In order to analyze the discrepancies on an individual level between self-reports and breath tests, categories of over-reports, congruent-reports and under-reports were derived for each of the estimates. Congruent reports were defined as agreement between BAC (as determined by the Alcolmeter) and EBAC (from clients’ reports) within ?r 0.05%. Discrepancies greater than 0.06% were defined as over-reports if EBACs were larger than BACs or under-reports if EBACs were smaller than BACs. The breakdown of cate- gories by estimate for the total sample is shown in Table 4. Note that the proportion of over-reporters is higher than the under-reporters for all esti- mates and is higher than the congruent reports for EBACl and EBACs. When

107

TABLE 4

Reporting status for each Estimated BAC for the total sample

EBAC1* EBAC2 EBACa EBAC, EBACF, % N % N % N % N % N _~ ___~

Over-report 56.9 37 47.7 31 41.5 27 27.7 18 23.1 15 Congruent** 36.9 24 43.1 28 47.7 31 60.0 39 58.4 38 Under-report 6.2 4 9.2 6 10.8 7 12.3 8 18.5 12

Total 100% 65 100% 65 100% 65 100% 65 100% 65

*EBAC+ is based on 0.010% BAC reduction per hour. EBACz is based on 0.015% BAC reduction per hour. EBAC3 is based on 0.020% BAC reduction per hour. EBACl is based on 0.025% BAC reduction per hour. EBACE, is based on 0.030% BAC reduction per hour.

**Congruent is defined as + 0.05.

those with zero BACs are eliminated from the analysis, the contrast becomes even more striking (see Table 5). In general, the proportion of under- and over-reporters is larger for this group when compared with the total sample. Moreover, over-reporters not only represented a greater percentage of the sample than under-reporters for all estimates (except EBACa which has the largest elimination rate), they also constituted a greater proportion of re- spondents than the congruent reports for all EBAC categories.

TABLE 5

Reporting status for each Estimated BAC for clients with positive BACs

EBAC1* EBACl EBAC3 EBAQ EBAC, % N % N % N % N % N

Over-report 74.1 20 66.7 18 63.0 17 37.0 10 29.6 8 Congruent* * 11.1 3 11.1 3 11.1 3 33.3 9 25.9 7 Under-report 14.8 4 22.2 6 25.9 7 29.6 8 44.4 12

-___

Total 100% 27 100% 27 100% 27 99.9% 27 99.9% 27

*EBAC, is based on 0.010% BAC reduction per hour. EBACz is based on 0.015% BAC reduction per hour. EBACa is based on 0.020% BAC reduction per hour. EBAC4 is based on 0.025% BAC reduction per hour. EBACE, is based on 0.030% BAC reduction per hour.

**Congruent is defined as * 0.05.

Correlations between BACs and EBACs for the total sample and for the subsample of people with, positive BACs reflect the lack of congruency in the latter group (Table 6). For the total sample all EBACs are correlated significantly with BAC. However, the opposite result occurs when those

108

TABLE 6

Correlations of BAC with Estimated BACs for total sample and for clients with positive BACs

Total sample N=65

Respondents with positive BACs N=27

r P ; P

EBAC1* 0.440 0.001 0.240 0.113 NS EBACz 0.429 0.001 0.246 0.108 NS EBAC, 0.411 0.001 0.238 0.116 NS EBAC4 0.385 0.001 0.232 0.122 NS EBAC5 0.347 0.002 0.208 0.150 NS

*EBAC1 is based on 0.010% BAC reduction per hour. EBACz is based on 0.015% BAC reduction per hour. EBACa is based on 0.020% BAC reduction per hour. EBAC4 is based on 0.025% BAC reduction per hour. EBAC5 is based on 0.030% BAC reduction per hour.

NS = not significant.

clients who have not been drinking are eliminated from the analysis; none of the correlations is significant. Thus it appears that the high proportion of respondents who had not been drinking (as measured by the BAC) greatly influenced the correlation coefficient but even more importantly, the amount of alcohol in the body (as measured by the BAC) is not related to the amount of alcohol reportedly consumed.

Discussion

The data just presented clearly differ from previous validity studies of self-reports of alcohol consumption. Thus it is extremely important to seek explanations of why such a disproportionate amount of over-reports oc- curred in this sample.

One area which may explain these unusual findings is the measurement instrument and the criterion. Unlike the follow-up study of Polich et al. [ 81 and the Sobell studies [ 21, a very specific time structure was imposed on the consumption items. Thus, respondents were asked to think of their recent drinking in one-hour segments (for most recent drinking) and four-hour segments (for less recent drinking). These separate timeframes may have induced respondents to offer an appropriate positive response for more time frames than were necessary, thus over-reporting consumption. Overlapping drinking time segments might have been perceived by the respondent as two separate time frames. Another aspect relating to the interview instrument which could explain these discrepancies is that no data were obtained during the interview on whether the respondent vomited during his/her last drinking

episode (if in the last 24 hours). It is not known if this is an appropriate explanation for the discrepant findings since other studies in this area have also not collected this information.

The criterion used to validate the responses (the Alcolmeter) may be another source of error which could explain these results. It is highly un- likely that the instrumentation itself could have accounted for such large discrepancies. The Alcolmeters were calibrated weekly and were tested against very reliable instruments used by the police (Breathalyzers) which, by law, are calibrated after every breath test. The readings could have been biased by recording errors since they were directly read from a drum on the front of the instrument by the interviewer. Yet, since most of the breath tests were done by one interviewer, it is less likely that this type of error would have a large effect on the findings.

A third area which may offer an explanation involves the sample popu- lation and the interview situation. It is quite possible that there was some unique factor about the new entrants into the program. Moreover the re- spondents in this study may indeed have behaved very differently from the follow-up respondents used by Polich et al. [ 81. However, this does not explain how they differed from the new entrants in the Sobell studies [ 21. The interview situation offers a very plausible reason why so many over- reports occurred. The respondents may have perceived their responses as having a major effect on whether they would gain entrance into a treatment program. As Robins suggests [ 91, the client has much “riding on” the out- come of the interview and when this occurs, the client is more likely to adjust answers in line with what he/she feels are the treatment personnel’s expectations. This over-reporting of alcohol consumption in a treatment interview situation (an amount which in some cases is physiologically impos- sible and would result in death) is also mentioned by Pernanen [lo] as a personal communication from W. Schmidt. Unfortunately, this phenomenon has not been fully explored in the literature.

The implications of this type of behavior are multifold. For clinicians, the knowledge of the patient’s actual BAC is, in itself, useful for ensuring accurate diagnoses and appropriate referral and treatment decisions [ll, 121 . In addition, knowledge of the congruency between reported recent alcohol consumption and actual alcohol consumption at the initial interview may relate to subsequent behavior in the program. Midanik [13] has reported that for emergency room entrants into a comprehensive alcoholism treat- ment program, under-reporters tended to remain in a residential treatment facility for the 2%day program, while the over-reporters were more likely to drop out permanently.

For program evaluators, this phenomenon of over-reporting upon en- trance to a treatment program casts some suspicion on follow-up measures which use reduction of alcohol consumption as a measure of improvement. Clients who over-report their consumption initially and then accurately report (or under-report) their drinking at the time of follow-up would inflate change rates and thus bias the findings in the direction of increased success rates.

110

In summary, data were presented which contradict most of the validity literature and the expectations of many clinic personnel that “alcoholics deny”. Following the “more is better” notion for reporting of alcohol- related behavior [ 11, over-reporting as a viable behavior is considered some- what aberrant and is virtually ignored. It is hoped that the data presented here will motivate researchers and clinicians to take notice of this phenom- enon and investigate this behavior in more depth in future studies.

Acknowledgments

This research was supported by an Alcohol Research Center Grant awarded by NIAAA and a Biomedical Research Support Grant awarded by NIH. The author wishes to thank Robin Room for his helpful comments, and Carol Seiden and Barbara Trite1 for their preparation of this manuscript.

References

1

2

3

8

9 10

11

12

13

L. Midanik, The validity of self-reports of alcohol consumption and alcohol problems: a literature review. &it. J. Addict., (in press). M. Sobell, L. Sobell and R. Vanderspek, Relationships among clinical judgment, self- report, and breath analysis measures of intoxication in alcoholics. J. Consul. Clin. Psych., 47 (1979) 204 - 206. L. Midanik, Early dropout in two sites of a comprehensive alcoholism treatment program. Unpublished doctoral dissertation, The Johns Hopkins University, Baltimore, MD, 1979. Intoximeters, Inc., Operator’s Manual distributed by Intoximeters, Inc., St. Louis, MO, 1976. U.S. Department of Transportation, Basic Training Program for Breath Examiner Specialist-Student Study Guide, U. S. Government Printing Office, Washington, DC. Maryland State Police, Chemical test for alcohol-breathalyzer operator’s program. Department of Post-Mortem Examiners, Office of the Chief Medical Examiner, 1978. M. Bogusz, J. Path and W. Stasko, Comparative studies on the rate of ethanol elimina- tion in acute poisoning and in controlled conditions. J. Forensic Sci., 2 (1977) 446 - 451. J. M. Polich, D. Armor and H. B. Braiker, The Course of Alcoholism: Four Years After Treatment, The Rand Corporation, Santa Monica, 1979. L. Robins, Deviant Children Grown Up, Williams and Wilkens, Baltimore, 1966. K. Pernanen, Validity of survey data on alcohol use. In R. J. Gibbens et al. (eds.), Research Advances in Alcohol and Drug Problems Volume 1. John Wiley, New York, 1974, pp. 335 - 374. J. H. W. Birrell, Clinical value of the chemical test for alcohol. Med. J. Austr., 3 (1967) 346 - 347. J. McMahon, The usefulness of a hand-held breath analyzer in the diagnosis and clin- ical management of patients in an ambulatory treatment program. In F. A. Seixas (ed.), Currents in Alcoholism. Volume 1. Biological, Biochemical and Clinical Studies, Grune and Stratton, New York, 1977, pp. 473 - 480. L. Midanik, Perceptual variables as factors in dropout from alcoholism treatment. Paper presented at the 11th Annual Medical-Scientific Conference of the National Council on Alcoholism/American Medical Society on Alcoholism/Research Society on Alcoholism, Seattle, Washington, May, 1980.