3.3. Risk of Bias in Studies
In terms of the overall risk of bias, there were concerns about the risk of bias for the majority of studies (14/15), with two of these assessed as having a moderate risk of bias and others having a high risk. A text summary is provided below for each of the six individual components of the risk of bias assessment. Risk of bias analysis was conducted and reported for each outcome.
We first analyzed the problem of selective reporting of information by observing which study problems were identified with the reporting of information in general or with the outcomes.
Following the Cochrane risk of bias assessment guidelines, if fewer than 95% of the randomized subjects are present in the results and therefore analyzed for the entire duration of treatment, regardless of whether they have discontinued treatment or not, the study is considered to have a high risk of bias. Many studies in our meta-analysis shared this unfortunate characteristic, as the study population is known for its tendency to drop out of treatment [
43] and to suddenly discontinue activities that have already been started (referred to D3 in
Table 1,
Table 2 and
Table 3). This does not justify the lack of data reported in the studies, but it allows us to understand why such a large number of studies fall into this bias. The articles by Ito, Donovan and Hall [
35] Rounsaville et al. [
36], Sandahl et al. [
37], Woody et al. [
19], Ojehagen et al. [
38] and Hoyer et al. [
42] fall into this problem. In these cases, it was not possible to exclude that the results were influenced by these omissions and, in particular, with respect to the most severe conditions.
Domain 2 reports variable results, with a similar rate of high, some concern and low risk. This refers to the “definition of the intended intervention” and a lack of high evidence because the majority of experimental designs combined psychological with medical, pharmaceutic or other psychological treatments that could not always be under strict control during the time of the study; however, this is required to reach the good methodological standard.
Regarding the reporting of outcome measures, as evaluated in D4 and D5, the studies obtained positive results, with few assessments of “some concern” and a large amount of low risk of bias. In general, it was found that the measures reported in the results section were consistent with those declared in the methodological and protocol section and those were coherent with the constructs that have been measured. Except in one case [
14], in which the research objectives were specified in a preliminary article, and with which it was possible to check the actual presence of all the declared measures, for the other studies, this assessment had to be based on the information contained in the articles themselves.
The randomization process, as explored by domain 1, collected generally good results, but in some cases, this process was not well described and the evaluation fell into the inferior quality category as “some concern” or “low quality”, affecting the final evaluation.
3.4. Results of Individual Studies
Looking for “substance use” in the alcohol group (
Figure 2, Part A), data from six studies were gathered. Specifically, the effect variability between the studies was quite wide, even if only in two cases was this significant, and this showed interesting differences in favor of CBT or dynamic treatment. The studies of Ito, Donovan and Hall [
35] and Sandahl et al. [
37] shared the same control treatment, i.e., a treatment of a cognitive behavioral nature based on the model devised by Marlatt [
44,
45], called the Relapse Prevention model. Ojahagen [
38] also used a therapy within the CBT paradigm, but it was based on another model, namely the multimodal theory, described by Lazarus [
46], and in Nyhuis et al. [
36], another CBT-based treatment was found, the Combined Behavioral Intervention, as described by Longabaugh et al. [
47]. Instead, regarding treatment-as-usual, as described in Gregory et al. [
27,
34], there was not enough information to trace it back to a specific theory. In the literature, this type of intervention is generally described as a set of cognitive behavioral techniques and the 12-step program; however, a more in-depth analysis [
48] showed how the theoretical categorization of these procedures was much more complex and multi-component in clinical practice.
Based on the results obtained by Sandahl et al. [
37], it appears that the Relapse Prevention model of treatment reduces alcohol consumption more than dynamic treatment, and the difference is significant (G = 0.741,
p = 0.041, CI = 95%). However, these data are not supported by other studies. Looking for other types of treatment, whether they originate from the CBT paradigm or fall within the description of treatment-as-usual, no significant differences have been reported. In contrast with these results, Nyhuis et al. [
36] reported a slight but significant difference that promotes the dynamic treatment for the reduction of intake (
Figure 2, Part A); however, the effect size was relatively small, indicating only modest differences (G= −0.363,
p = 0.019, CI = 95%).
Both with regard to participation (
Figure 2, Part B) and other symptomatic conditions (
Figure 2, Part C), the individual studies did not report significant differences between interventions.
Concerning the use of cocaine (
Figure 3, Part A), it was possible to collect data on the trend in its use in patients under treatment for only two studies. Although two studies are sufficient to conduct a meta-analysis, this limits generalizations. The main difficulty that limited the number of eligible articles was the quality of the outcome report, often without traceable data for the meta-analysis.
Crits-Cristoph et al. [
14] compared the dynamic treatment to three different types of therapy, so the effect size is the average of the differences observed between these and the dynamic therapy used. Both in the papers of Crits-Cristoph et al. [
14] and Carrol, Rounsaville and Gawin [
39] no significant trends were found. In the first study, differences were not significant, except for “individual counseling”, which showed better, statistically significant results compared to “supportive expressive” dynamic therapy. In the second study, results were also not significant, and no evidence was found in favor of either (G = 0.753,
p = 0.059, CI = 95%).
Concerning participation in the treatment (
Figure 3, Part B), the differences in the comparisons made by Crits-Cristoph et al. [
14] and Carrol, Rounsaville and Gawin [
39] proved to be not significant. These results showed an effective equivalence between observed treatments. One interesting result that emerged from Crits-Cristoph et al. [
14] is the significant superiority of both dynamic and CBT interventions to individual and group counselling, in contrast with the result within the “substance use” outcome.
Concerning the “symptomatic condition” outcome, only Crits-Cristoph et al. [
14] reported data that could be used, while Carrol, Rounsaville and Gawin [
39] and the other articles did not consider this outcome or did not report the achieved results. From the work of Crits-Cristoph et al. [
14], it is possible to conclude that, besides the treatment of choice, the observed rate of psychological symptoms tends to decrease together with a decrease in substance use. These results are in line with those reported by NIMH on the co-occurrence of psychiatric disorders and substance addiction, showing how these often support each other [
49], and would confirm the results of analyses previously carried out on this topic with reference to other symptomatic conditions, such as depression [
50].
Concerning opioid use (
Figure 4, Part A), data were available from four studies, three of which were conducted by the same authors, one as a follow-up study and the other as a validation study. The three studies by Woody et al. show conflicting results; Woody et al. [
41] are reported here as Woody et al. [
51] as some data reported erroneously in 1983 have been highlighted and corrected later by the authors, so Woody et al. (1987) [
52] is reported as a follow-up study of corrected Woody et al. (1990) [
51]. Non-dynamic treatments used as comparisons, drug counseling and cognitive treatment, had no reported significant differences from the expressive supportive treatment. In the follow-up study [
52], on the other hand, two and a half years after the start of treatment, the non-dynamic treatments resulted in a significantly higher reduction in consumption than the supportive-expressive treatment. This difference was very large, in addition to being statistically significant (G = 1.051,
p = 0.005, CI = 95%). In the validation study by Woody et al. [
19], only drug counselling was considered as the control condition. In this case, as in the first study, the differences between treatments were minimal and not significant. In Rounsaville et al. [
36] instead, both interventions led to an increase in the use of substances over time, so it is possible to state that the two treatments proved to be equally ineffective in reducing opiate consumption.
Concerning treatment participation (
Figure 4, Part B), data from three studies were collected. In all three cases, whether they were alternative methadone treatment by Shaffer, LaSalvia and Stein [
18], drug counselling by Woody et al. [
19] or the low-contact treatment used in Rounsaville et al. [
40], no significant differences were recorded. The only exception was Woody et al. [
19], where a medium-sized statistically significant difference in favor of the control condition was reported (G = 0.497,
p = 0.0010, CI = 95%).
Regarding the symptomatic improvement associated with the treatment of opioid use (
Figure 4, Part C) differences favorable to one treatment could not be observed in both studies, i.e., Woody’s follow-up study [
52], which also reported the original data from Woody et al. [
41], and the 1995 validation study [
19] from the same authors. There they found no significant differences among dynamic treatment, drug counselling and cognitive treatment.