The parent–child relationship and adolescent alcohol use: a systematic review of longitudinal studies

An extensive search of the literature was conducted in order to identify longitudinal studies regarding the association between the PCR and adolescent alcohol use. The following databases were searched: Medline, PsycINFO, and EMBASE. The search was limited to studies which were published in the English language, during the period 1985-July 2011, and which focused on children or adolescents. The search strategy combined the following three sets of terms related to respectively alcohol use, PCR, and a longitudinal study design; (1) alcohol us* OR alcohol drink* OR alcohol dependen* OR alcohol abus* OR alcohol consum* OR binge drink* OR heavy drink* (textwords), OR alcohol drinking (Mesh-term); (2) parent* (textword), OR parenting OR parents OR parent–child relations OR family OR child rearing (Mesh-terms); (3) longitudinal OR cohort OR follow-up OR prospective OR baseline OR mixture/mixed/growth model* OR growth curve* OR generalised/generalized estimating/estimation equation* (textwords), OR GEE (title or abstract), OR longitudinal studies (MeSH-terms). If the keywords did not exist in PsycINFO or EMBASE, alternative keywords were identified in the index of the database in question.

For the selection of the studies, a list of inclusion and exclusion criteria was developed in order to detect relevant longitudinal studies. A study was included in the review if: (1) it was a prospective cohort study; (2) there were at least two assessments of alcohol use between ages 10–17 in order to quantify change in alcohol use; (3) the total follow-up period was at least one year; and (4) data on alcohol use were presented separately but not if it only presented these data for the combined use of several substances. For measurement of the PCR, we only included those studies which measured (aspects of) the PCR construct as defined in the introduction. Each possibly relevant study was checked on item-level to be sure this construct was measured. Some studies used questionnaires composed of items regarding the family in general along with items regarding the parents. These studies were included if over half of the items concerned the parents. Studies were excluded if they included only clinical populations and if alcohol use was not the outcome measurement. In addition, dissertation abstracts, reviews, comments, letters, and editorials were excluded.

The selection of the studies was performed by the three authors. First, the titles or abstracts of the identified references were screened by the first author (LV). Subsequently, for the remaining references, the full paper was retrieved and was screened regarding the selection criteria. In case of doubt, the study was discussed by the first (LV) and the second author (AFW) in order to reach consensus. If necessary, the third author (SAR) was consulted. Finally, the reference lists of all the selected publications and of relevant systematic reviews were screened for potential missing studies.

The methodological quality of the studies was assessed using a checklist derived from Hayden et al. [20]. This guideline lists criteria regarding six domains of potential biases in prognostic studies: (1) study participation; (2) study attrition; (3) predictor measurement; (4) outcome measurement; (5) confounding measurement; and (6) analysis. Sixteen relevant criteria considering each of the six potential biases were selected and adapted to the review question (Table 1). For the judgment of the validity and reliability of the measurement of alcohol use (criterion K), requirements concerning assessment of alcohol consumption were taken into account [21]. All the selected studies were independently assessed by the first two authors (LV and AFW). All criteria were rated as either “yes” (+), “partly” (±), “no” (−), or “unsure” (?) – and given 2, 1, 0, and 0 points, respectively – or as “not applicable.” In case of any disagreement, consensus was reached by discussion and, if necessary, the third author (SAR) was consulted. If ≤50% of the maximum score for a possible bias was obtained, the bias was scored as 1 (considerable risk for bias); if >50% of the maximum score was obtained, the bias was scored as 0 (low risk). For each study to obtain a total quality score, the numbers of the biases were summed, with the result ranging from a possible 0 to 6. A study was judged to be of high quality if there was a low risk for each domain of potential bias.

The heterogeneity of the included studies precluded a meta-analysis to summarize the results. Therefore a best evidence synthesis was applied to determine the strength of the evidence in regard to the effects of the PCR on the change in adolescent alcohol use, as used in other reviews [22, 23]. This synthesis was based on the number, the quality, and the outcome of the studies, and leading to four levels of strength for the evidence of the existence of an association between the PCR and alcohol use: (1) strong evidence: consistent findings in at least two high-quality studies; (2) moderate: consistent findings in one high-quality study and at least one low-quality study; (3) weak: findings in one high-quality study or consistent findings in at least three low-quality studies; and (4) inconclusive: inconsistent findings irrespective of study quality, or less than three low-quality studies available. Findings were considered to be consistent when at least 75% of the studies involved had agreed on the existence and direction of the association between the PCR and alcohol use.

The search strategy resulted in 2811 references of which 28 studies were included (for detailed information about any of the steps in the screening process, see Figure 1). Fifteen of these 28 studies were not included in the review of Ryan et al. [14], 27 studies not in the review of Foxcroft and Lowe [12], and none of the studies were included in the review of Vakahali [13].

The characteristics of the studies we selected are summarized in Table 2. Twenty-one studies were undertaken in the USA, two in the Netherlands, one in Finland, one in Spain, one in New Zealand, one in Sweden, and one in Taiwan. The studies included different age groups at baseline, which varied between age 10 to over 18. The number of participants in the studies ranged from 166 to 4731. The total follow-up period ranged from 1 to 12 years and the number of waves varied from 2 to 8.

The majority of the studies measured the PCR as protective or as risk factor. In seven studies both potential roles of the PCR were measured. In two studies, the quality of the relationship was based on parent report, while in all the other studies adolescent report was used. Table 2 shows all terms and descriptions that were used for the PCR in the included studies. In the current review, the term PCR refers to all those terms.

A large diversity of statistical approaches was used to analyze the association. Almost half of the studies made use of “traditional” methods as analysis of variance (ANOVA) [24, 25] or a form of regression analysis [26‐37]. Other studies used more “sophisticated” methods as generalized estimating equations (GEE) [38‐40], hierarchical linear modeling (HLM) [41, 42], structural equation modeling (SEM) [43‐49], latent class growth analysis (LCGA) [24], multiple mediation modeling [50] or a combination of these methods [51].

Some studies showed results that need an additional explanation. A number of the studies reported only one beta or odds ratio while the study had more than two waves [33, 35, 38, 39, 41, 43, 46, 50]. One of these studies used only two of the three available waves without a clear reason for this [33] whereas the other study did this because neither the PCR nor alcohol use was measured at the third wave [46]. For the other studies that reported only one beta or odds ratio but the study had more than two waves, the data analytic approach provided an explanation for this. Chuang et al. [43] and Latendresse et al. [50] measured the PCR at one of the first two waves (T2 and T1, respectively) and alcohol use at the remaining two waves. In their analyses alcohol use at the last wave was predicted by the PCR, controlling for alcohol use measured at the other wave at which it was assessed in that study (T1 and T2, respectively). Andrews et al. [38] and Droomers et al. [39] did a GEE-analysis. In these GEE-analyses all waves were combined, leading to one estimate of the effect of the dependent variable on the independent variable and thus to only one beta [38] or odds ratio [39]. Kosterman et al. [35] used survival analysis which yielded only one beta based on the eight waves. Aseltine and Gore [41], and Gutman et al. [42] used HLM-analyses which were equivalent to the LGCA-analyses as done by Barnes et al. [51]. In all three studies latent growth curves were estimated resulting in one intercept and one slope for the PCR was related.

Also three other studies need an additional explanation. Wu et al. [25] used ANOVA for which F-values were not reported but p-values were given. Flory et al. [24] used LCGA with ANOVA. First a LCGA analysis was done which resulted in the identification of three subgroups: early onset, late onset, and non-users. Next ANOVA was done to test differences between the subgroups with respect to PCR. Although the focus of this systematic review was on the direct effect of the PCR on alcohol use, Barnes et al. [51] found an indirect of the PCR on alcohol misuse that operated via parental monitoring.

Results of the quality assessment are shown in Table 3. Nine studies (32%) met the criteria of each of the domains of potential biases and therefore were judged as high quality. The remaining 19 studies (68%) were at risk of one or more biases.

There were notable limitations concerning study attrition. More than half of the studies (57%) did not have an adequate response rate, or they had important differences between participants and dropouts. Twenty-one percent of the studies were at risk of bias on the measurement of the outcome and also 21% of the studies were at risk of bias on study participation. Finally, a bias, whether on the predictor measurement, on the confounding measurement, or in the analysis, was reported in only 7% of the studies. In total, 448 items were scored twice, which resulted in disagreement on 67 items (15%), largely caused by reading errors. In all cases of disagreements, the two assessors could reach consensus during discussion.

The effect of the PCR on alcohol use found in the studies included is reported in Table 2. Five studies [26, 29, 34, 39, 42], of which none was rated as high-quality, reported statistically significant negative associations between the quality of the PCR and alcohol use for the whole group. However, the findings in some of these studies [29, 42] were equivocal. Crawford and Novak [29] only found a significant association for onset of heavy drinking but not for onset of alcohol use, number of drinks in lifetime, or times heavy drink. Gutman et al. [42] found significant associations in only three of four models. Those models studied the short and long-term effects of positive identification and negative interactions on alcohol use. The long-term influence of negative family interactions on alcohol use was not significant, but the remaining three associations studied were.

Seven studies [28, 38, 40, 41, 46, 48, 49] also found a negative association between the PCR and alcohol use, although only for specific groups. Two of these seven studies only found an effect for a certain gender [38, 48], another three only for certain age groups [28, 41], another two only for a certain gender with a certain age [40, 49]. However, these studies did not show consistent results regarding the effect of gender or age on the association between the PCR and alcohol use. In the last study in which a subgroup analysis was performed, participants were split into a low and high-quality group for the PCR [46]. This study showed that only those who started late drinking alcohol and had a high-quality relationship with their parents consumed little alcohol later on. Of all these studies three were classified as high quality [46, 48, 49].

The remaining sixteen studies [24, 25, 27, 30‐33, 35‐37, 43‐45, 47, 50, 51], of which six were rated as high quality [25, 30, 37, 47, 50, 51], did not find a significant association in the whole group nor in a subgroup. Three of these studies [30, 31, 45] performed a subgroup analysis by gender or by age but did not find significant associations in one of these groups. The other thirteen studies examined the group as a whole and did not find any significant effects.

Table 4 summarizes the level of evidence for the PCR as protective (e.g., warmth, support) and as risk factor (e.g., conflict, rejection) regarding various types of alcohol behavior. We found weak evidence for the conclusion that there is an association between the PCR and frequency and/or amount of use. No systematic differences existed in the level of evidence by types of alcohol behavior. Overall, the level of evidence for the existence of an association between PCR and alcohol use seems to be stronger for PCR as protective than as risk factor.

According to the criteria of the best-evidence synthesis method, there is weak evidence for an effect from the PCR on change in adolescent alcohol use. The evidence was classified as weak because less than 75% of the studies agreed on the existence and direction of the relation between the PCR and alcohol use.

The effect of the PCR as protective and as risk factor on the different alcohol behaviors is shown in Table 4. The strength of the evidence for each relationship is weak to inconclusive. Overall, the level of evidence for an effect of the PCR on adolescent alcohol use is stronger for protective factors than for risk factors.

Strengths of this review are the systematic evaluation of the evidence taking into account the methodological quality of individual studies and the consistency of the research findings. Another strength is the limited influence of publication bias. In our review we included studies reporting the association of the PCR with alcohol use although this was not the main research question. With this inclusion also non-significant results were taken into account. Despite of the comprehensive search strategy, a limitation is that we might have missed some studies due to the restriction imposed by using only those studies published in English. However, most reviews use a language restriction and no evidence was found that such a restriction leads to a bias [59]. Furthermore, due to restriction of the search to electronic databases and to year of publication, we might have missed some studies. But it is unlikely that unpublished studies, non-journal studies, or studies published before 1985 would influence our conclusion regarding the level of evidence. Besides that, the likelihood that we have missed some studies is small because we screened the reference lists of the selected studies and of relevant systematic reviews. Another limitation of our study may be the use of a cut-off point for the identification of high-quality studies which we used to determine the strength of the evidence. A low risk for a bias was supposed if >50% of the maximum score for that bias was obtained; this resulted in nine high-quality studies. With a cut-off point of ≥50% sixteen studies would have been rated as high quality. Nevertheless, due to the lack of consistent results (less than 75% of the studies found significant associations), the conclusion regarding the level of evidence would end up being the same.

We found weak evidence for a causal association between the PCR and alcohol use. The causality of this association and, if causal, its direction is far from clear. It might also be for example that the PCR is only causally related to certain types of alcohol behaviors. Our study found stronger evidence for an association of the PCR with frequency and/or amount of use than with other types of alcohol use. However, research that studied the association for several types of alcohol behaviors, so the comparison of effects is more reliable, is lacking. Additional research may clarify this. Furthermore, additional research should focus in particular on the role of overprotection by parents regarding the relation between the PCR and alcohol use as discussed above. Considering the gap in the literature, this research should also concern the extent of parental influence across several stages of alcohol use, and the additional effects of peer behavior. Any future research should further be designed in such a way that it also enables the assessment of potential reversed causality. Furthermore, for progress in this area more standardization in the measurement of both PCR and alcohol use, the follow-up periods, and analytic procedures is needed, considering the significant heterogeneity present in our review.