All 60 perpetrators of IPV met DSM-IV diagnostic criteria for alcohol dependence. Four alcoholic perpetrators were missing the baseline IS and were dropped from analysis. Thirty-two alcoholic perpetrators had unexcused missed assessments. Ten of these 32 perpetrators had stopped their medication for prolonged periods of time, had missed multiple assessments, and were readmitted to the hospital as described under the Study Design section. The remaining 22 alcoholic perpetrators failed to complete the 12 weeks of the study and were completely lost to follow-up ().
The 10 readmitted perpetrators were eliminated from all analyses except the intent-to-treat analysis because of the possible confound of being off medication treatment for a prolonged period of time and the additional treatment that they received. There were no baseline IS (F1,59
= 1.43, P
= .24) or alcohol lifetime consumption (F1,59
= 0.01, P
= .92) differences between these 10 perpetrators and those retained for analysis. All analyses were performed using STATISTICA, version 7.1 (StatSoft, Inc, Tulsa, Oklahoma).31
shows the lifetime characteristics of all 60 alcoholic perpetrators. There were no significant differences between the treatment groups. shows the number of missing perpetrators at weeks 4, 8, and 12 as a function of fluoxetine versus placebo for the 22 lost-to-follow-up perpetrators.
Lifetime Characteristics of 60 Alcohol-Dependent Perpetrators of Intimate Partner Violencea
Number of Perpetrators Who Failed to Complete 4, 8, and 12 Weeks of Treatment
Irritability Subscale of the Modified Overt Aggression Scale
The IS score was analyzed using 3 different statistical methods. These methods utilized the completer analysis, the last-observation-carried-forward analysis, and the intent-to-treat analysis.
Completer data analysis, utilizing scores from baseline through week 12, was a repeated-measures analysis of covariance with the IS score as the repeated measure and the baseline IS score as the covariate. The between-groups measure was the drug factor. There was no significant repeated-measures effect (F2,42 = 0.85, P = .43) or interaction effect (F2,42 = 0.02, P = .98). There was a significant drug effect (F1,21 = 12.09, P = .002) ().
Effect on the IS Score of the MOAS for Perpetrators Who Completed 12 Weeks of Treatmenta
The last-observation-carried-forward-analysis, in which the last observations from week 4 and week 8 were carried forward to week 12, was utilized to account for missing data. We performed an analysis of covariance with the IS score as the dependent variable at week 12 and the IS score at baseline as the covariate. The drug factor was the between-groups variable. There was a significant drug effect (F1,32 = 4.24, P = .048).
To examine the possibility that the missing IS scores were missing completely at random, we performed the following analyses. Using an analysis of variance model with a drug factor, with a missing value factor, and with the IS score at baseline as the dependent variable, it was found that the missing value at week 4 could not be predicted from the data at baseline (F1,42 = 0.10, P = .75). A multivariate test was performed for missing data at week 8 using data from baseline and week 4. The missing data at week 8 could not be predicted from values of the observed data at baseline and week 4 (F2,29 = 0.34, P = .71). A multivariate test was performed for missing data at week 12 using data from baseline, week 4, and week 8. The missing data at week 12 could not be predicted from values of the observed data at baseline, week 4, and week 8 (F3,22 = 0.66, P = .59). We conclude that, with respect to the IS score, the data may be missing completely at random.
To examine the possibility that the missing data were related to alcoholic drinking, we performed an analysis of variance with total lifetime alcohol consumption as the dependent variable and the missing data factor at week 4 and found a trend for missing data at week 4 (F1,40 = 3.83, P = .057). This finding suggests that alcohol may be a contributing cause of the missing data. There was no significant correlation between total lifetime drinking and baseline IS scores, further suggesting that the missing data mechanism is unrelated to the IS score. Comparison of fluoxetine versus placebo for missing data was not significant (χ22 = 1.68, P = .43) (). These results provide some evidence that the missing IS scores may be considered as missing completely at random.
All 60 alcoholic perpetrators were entered into an intent-to-treat analysis utilizing scores from baseline through week 12. A repeated-measures analysis of covariance was performed with the IS score as the repeated measure and the baseline IS score as the covariate. The between-groups measure was the drug factor. There was no significant repeated-measures effect (F2,54 = 0.70, P = .50) or interaction effect (F2,54 = 0.04, P = .96). There was a significant drug effect (F1,54 = 5.0, P = .034).
Spielberger State Anxiety Inventory and Hamilton Depression Rating Scale
To examine whether depression and anxiety could have influenced the drug effect on the IS score, we performed a repeated-measures analysis of covariance for the Hamilton Depression Rating Scale scores and Spielberger State Anxiety Inventory scores (). Using the baseline measures as covariates, there was no significant drug effect, repeated-measures effect, or interaction effect.
Behavioral Ratings (mean ± SD) During 12 Weeks of Fluoxetine or Placebo Treatment
All perpetrators randomly assigned to receive fluoxetine had measurable drug levels with the following group means ± SDs for fluoxetine plus its metabolite norfluoxetine (ng/mL): week 4 = 129.8 ± 59.1 ng/mL, week 8 = 181.5 ± 90.4 ng/mL, and week 12 = 212.3 ± 116.1 ng/mL.
Spouse and Significant Other Evaluation
A repeated-measures analysis of variance for time points baseline and week 12 was performed with drug treatment as the between-groups factor. Both the nonphysical and physical Partner Abuse Scale ratings showed a significant time effect (F1,11 = 24.2, P = .0005 and F1,11 = 10.2, P = .009, respectively), with no significant interaction effect and no significant drug effect ().
Scores on Partner Abuse Scale Completed by Spouses/Significant Othersa
Safety and Tolerability
Fluoxetine was well tolerated. Only 2 perpetrators were maintained on less than 40 mg/d of fluoxetine. There were no serious adverse events.