Commentary on "high-risk behavior in patients with alcohol dependence".

Sir,

I read with interest the study on high-risk behavior in patients with alcohol dependence by Korlakunta and Reddy[1] in March–April issue of 2019. The study used the high-risk behavior questionnaire to estimate the rate of such behavior in patients with alcohol dependence. It was found that 120/200 (60%) patients had one high-risk behavior associated with alcohol use. The conclusion from the study was that “the occurrence of high-risk behavior was substantial among patients with alcohol dependence syndrome.” Furthermore, several demographic and clinical factors were shown to be associated with high-risk behaviors. However, there are several glaring errors in the manuscript.

The objective of the study was mentioned as “to assess the relationship between high-risk behavior and alcohol dependence.” In the absence of a control group in the current study, concluding such a relationship is not possible. The period for the assessment is not clearly stated. Assuming the occurrence of such behavior any time during their drinking career, it is difficult to fathom that each of 120 individuals had only one high-risk behavior. More likely, there would be multiple such behaviors in an individual subject; the recall of such events, however, could be a source of bias. Surprisingly, there is no mention of the eligibility criteria of the study participants, except for International Classification of Disease-10 criteria of alcohol dependence syndrome.

The manuscript mentions the use of “Pearson's correlation coefficient, t-test, and logistic regression” under the methods section. However, the only analysis that is reported in the original manuscript is the “chi-square test” in Tables 3 to 5. Furthermore, the Chi-square test has been used for variables where cells are empty, which is a violation of assumption for the test. One of the assumptions states that the expected frequency in the cell should be five or more in at least 80% of the cells, and no cell should have an expected count of <1.[2] If the assumption is violated, an alternate test such as Fisher's exact test should be performed.[2]

The researchers have created meaningless categories out of continuous variables such as age, age at initiation of drink, age at dependence, and duration of dependence, which limits interpretability. Specifically, such categorization leads to loss of power in detecting associations and residual confounding, hence, better avoided.[3] Multiple hypothesis testing was carried out without any corrections, which could inflate the family-wise error rate. Reporting of confidence interval could have been more meaningful.

Several other minor errors in reporting could have been avoided. For example, “P = 0.000” should be mentioned as “P < 0.001,” as P value can never be zero. Uniform use of three digits after decimal points for P values and two digits after decimal points for the rest of data would enhance readability.

Financial support and sponsorship

Nil.

Conflicts of interest

There are no conflicts of interest.