DISCUSSION

The results of this study indicate that previous long-term benzodiazepine users, assessed after at least 6 months of abstinence, continue to display cognitive deficits in a number of areas, compared to matched controls. Significant, moderate-to-large effect sizes were observed within the areas of verbal memory, motor control/performance, and nonverbal memory when comparing the BZD and NML groups. Significant differences were also found between comparison of the BZD and ANX groups on these measures.

The observation that the ANX and NML performed similarly on most measures and significantly better that the BZD groups implicates long-term benzodiazepine use as the most plausible explanation for the difference. Elevated anxiety levels could not account for poorer performances observed by the BZD group, as the both BZD and ANX groups displayed very similar State Anxiety levels. Gorenstein et al. (1995) also found their group of previous long-term benzodiazepine users, assessed at long-term discontinuation (average 10 months), to be impaired across a number of cognitive areas, including verbal memory, when compared to both an anxious control group and a normal control group. Furthermore, the past benzodiazepine user group and anxious control group did not differ significantly on anxiety scores, which were also significantly different from the normal control group. These authors argued that the absence of significant differences between the performance of anxious benzodiazepine-free controls and normal controls on all measures (except a manual dexterity task) indicates that poor performance in the past benzodiazepine user group could not be attributed to anxiety (Gorenstein et al., 1995). Similarly, other researchers have discounted anxiety as the likely cause of poor performance in long-term benzodiazepine users when the effects of anxiety are accounted for (Golombok et al., 1988).

On only one measure, the Austin Maze (nonverbal memory), did the ANX and NML groups differ. In addition, on this measure there was no difference between the BZD and the ANX groups, indicating that elevated anxiety levels may have impacted negatively on performance. Qualitatively this was certainly the case, with many of the participants in the both the BZD and the ANX groups reporting that this task, which includes a buzzer to indicate an incorrect move, was particularly anxiety provoking.

While the overall results of the current study are difficult to compare to previous studies due to the variety of study designs and cognitive areas examined, support for impairments in the areas assessed in this study is provided by the following studies: verbal memory (Tata et al., 1994), motor control/performance (Gorenstein et al., 1995; Petursson et al., 1983), nonverbal memory (Birzele, 1992), visuospatial skills (Tata et al., 1994; Golombok et al., 1988; Sakol & Power, 1988), and attention/concentration (Birzele, 1992; Golombok, 1989; Sakol & Power, 1988). The current study did not detect significant differences in performance in the areas of visuospatial skills and attention/concentration between the BZD group and controls. However, despite significance levels failing to reach the Bonferroni-corrected p value of .002, the observed trends indicated that the BZD group performed poorer than the controls in these areas, with significance levels consistently less than .05.

The findings of this study are supported by a comparison of the areas of impairment found to those in the previous meta-analysis (Barker et al., 2004b) by arranging the areas investigated in order of most to least impairment indicated by effect size magnitude (Table 4). The previous meta-analysis integrated the results from a number of studies that reassessed withdrawn long-term benzodiazepine users after at least 6 months of abstinence. By comparing the BZD group to the combined control group, calculating a mean effect size for each cognitive area, and then ordering the effect sizes from greatest to smallest, a very similar pattern is evident. For example, large differences were consistently observed in performance between groups on verbal memory tasks in both the current study and the previous studies included in the meta-analysis, resulting in the largest effect size observed in the cognitive area of verbal memory.

The effect sizes observed in this study were larger in magnitude to those found in the previous meta-analyses for all categories except verbal memory. A possible explanation for this is the high level of control and the choice of dependent measures. The tests used in this study were chosen based on their sensitivity to detect deficits in particular areas, whereas the category results form the previous meta-analysis are based on results from a variety of measures. The similar effect size magnitude found for the category of verbal memory does not follow this argument but may either reflect a ceiling effect, or be due to the minimal variation of measures used to assess verbal memory. Further investigation revealed that, of the studies in the previous meta-analysis that investigated verbal memory, most used a very similar type of story memory task to each other and to the current study.

The results of this study, combined with the results from the previous meta-analyses indicate that, in some areas, differences do exist between the performance of previous long-term benzodiazepine users and those who have not used this medication. The failure of some previous studies to detect differences may again be attributable to the lower level of control and the small sample sizes often employed, resulting in the small effect sizes and nonsignificant differences reported. However, by combining all of the positive and negative effect sizes from the available previous research, the magnitude of the resulting category effect sizes support the existence of differences between previous long-term benzodiazepine users and controls (Barker et al., 2004a, 2004b) .

Tata et al. (1994) have previously pointed out the well-established link between acute benzodiazepine administration and adverse effects on memory and arousal, and attempted to link this to the high density of benzodiazepine receptors found in the hippocampus and reticular formation (Wolkowitz et al., 1987). The two largest differences observed in the current study were in the areas of verbal memory and motor control/performance—functions largely subserved by the hippocampus and reticular formation. However, further research is required before any conclusions may be drawn on the mechanism of persisting deficits, or the relationships between the pattern of deficits observed after withdrawal and physiological differences in the distribution of benzodiazepine receptors in the brain.

While this paper addresses some of the methodological criticisms of previous research in this area, a number of limitations remain. First, the small sample size of twenty in each group resulted in limited power to detect differences in some of the areas assessed. Furthermore, the Bonferroni correction applied, due to the number of analyses conducted, resulted in an extremely conservative test of significance. As a result, some real differences may have not been detected. The small sample size also precluded a comprehensive regression analysis of the contribution of length of use, dosage, and length of abstinence. This type of analysis is the logical and necessary next step in the body of research in this area.

The small sample sizes in this study are partially attributable to the method of recruitment and the strict exclusion criteria that were applied. In the process of attempting to address one of the previous criticisms—that some studies tend to include patients with high alcohol or current antidepressant use—the difficulty of finding previous long-term benzodiazepine users not currently taking other medication became apparent. It seems that a common practice for long-term benzodiazepine users who wish to withdraw is to replace their benzodiazepine medication with antidepressant medication. Many of those excluded from the study who had withdrawn, experienced a return of their initial symptoms and were prescribed a selective serotonin reuptake inhibitor (SSRI) to treat their anxiety. The increasingly popular SSRIs were often considered the lesser of two evils by both patients and their medical professionals.

Second, the recruitment method employed may have resulted in a biased sample. Most of the patients who attend TRANX do so to gain assistance in withdrawing from their medication. However, when mailing out to past TRANX clients, it was not known whether they had successfully withdrawn, their current withdrawal status, their current medication regime, or whether they had even been taking a benzodiazepine medication. This would have resulted in the invitation letters being sent to a number of people who were actually not eligible to participate. Of those who responded to the mailout, less than half were eligible for a variety of reasons. The second method of recruitment involved a newspaper advertisement. Consequently, the resulting group who were recruited into the study may have been a subgroup of previous users who were experiencing cognitive problems and were therefore more likely to respond to the letter or advertisement in order to undergo some investigation of their difficulties. The majority of the enrolled participants cited concerns about health or side effects as the predominant reason for wanting to discontinue. Furthermore, it may be the case that the patients who successfully withdraw from benzodiazepine medication do so because they are insightful of cognitive impairments, while those not experiencing any negative effects are less inclined to attempt to withdraw.

Third, it is not possible to comment on the preexisting cognitive functioning of the past benzodiazepine users. Lucki et al. (1986) have previously pointed out the difficulty that exists in estimating what a participant's performance may have been prior to, or without, long-term benzodiazepine use. The present study used an estimate of premorbid intelligence; however, premorbid performance on measures such as memory or psychomotor performance is impossible to determine. While the groups in the present study did not differ statistically on the measures of premorbid intelligence (NART–II) and general intelligence (WASI FSIQ), these differences almost reached significance and should be acknowledged. Despite controlling for education, the NML group appeared to obtain a higher WASI FSIQ than both the BZD and ANX groups. In addition, the ANX group appeared to achieve a higher NART–II score than both the BZD and NML groups. These two measures would be expected to correlate highly and therefore this finding is considered highly unusual. One possible explanation is that the NART–II predicts WAIS–R FSIQ whereas the WASI is an estimate of WAIS–III FSIQ, with the older normative data tending to yield comparatively higher IQ scores.

Fourth, while differences between groups in the category attention/concentration did not reach significance, the use of a measure more sensitive to attentional dysfunction than the Trail Making Test–Part A would have been more appropriate, and may have revealed some differences.

Fifth, toxicology testing was not available during this study, and therefore the researchers relied on self-reported status of continued abstinence, and current use of illicit drugs or alcohol. While there was no way to confirm these states, there was no incentive to be other than truthful and the research was conducted completely independently of any connections the participant may have had to treatment facilities.

Finally, the same examiner conducted all of the psychometric testing, and due to the study design, was also involved in recruitment and therefore not blind to participant's group membership. While the possibility of bias should be considered, all of the tests used involved standardized administration and scoring procedures, and would not be expected to be affected.

This study attempted to address some of the problems inherent in the research in this area by conducting a small well-controlled study, including an anxious control group, excluding patients with illicit drug or high alcohol use, and excluding patients in any group who take antidepressant or other psychotropic medication. Furthermore, we present the results in a comprehensive manner that is amenable to inclusion in future meta-analyses. As a result of these strict inclusion criteria, the BZD group and the nonmedicated ANX control group proved to be very difficult to recruit and may have introduced the type of selection bias previously discussed. Recruitment difficulties also resulted in a smaller than anticipated sample size. Irrespective of these difficulties, the authors maintain that this approach to furthering our understanding of the nature of effects caused by long-term use of benzodiazepines is the most appropriate.

This study provides additional support for the hypothesis that long-term benzodiazepine impacts negatively on cognitive functioning in a number of areas. Given that the length of abstinence from benzodiazepines in this study was between 1 and 29 years, the persisting cognitive effects observed indicate that the impairments may also be long-term. These findings have important implications for those considering taking benzodiazepine medication, as well as those considering prescribing these drugs. Informed decisions regarding this therapy should be only made after considering the potential long-term impacts in conjunction with proposed treatment benefits.