Participants were recruited between April and June 2016 for baseline (T1) interviews, and contacted again between April and June 2018 for 24-month follow-up interviews. At baseline, surveys were collected from 1072 of 1351 eligible adolescent wives (79.3% female participation), 968 of whom provided survey data at follow-up (90.3% female retention); 1080 of 1351 eligible husbands completed surveys at baseline (79.9% male participation), of whom 773 participated in data collection at follow-up (71.6% male retention) (Fig. 1). Additionally, 27 women who did not provide surveys at baseline but provided surveys at follow-up were added to the sample. In total, 1099 women were included in the final sample. Intervention delivery and 24-month follow-up data collection were completed as outlined in the protocol paper [18].

Average age of wives at baseline was 17.3 years [standard deviation (SD) 1.5 years], and average age of husbands at baseline was 25.6 years (SD 5.3 years) (Table 1). Wives were, on average, 14.2 years old at marriage (SD 1.9 years) (Table 1). Education was low for both women and men, with limited levels of Quranic school (16% women, 20% men) and any modern school (35% women, 47% men), and high levels of no schooling (48% women, 30% men). At baseline, 40% of adolescent wives had never given birth, and most husbands (84%) had only one wife. Few wives had travelled outside the village for more than 3 months in the past year (6%), but this was common for husbands (67%). Of six assets assessed in the baseline survey, households owned on average 2.1 (SD 1.2 assets).

Treatment and control arms were not entirely equivalent at baseline. Husbands in the control villages were more likely to have no schooling, to have spent more than 3 months away from the village in the past year, and to have a higher average number of assets than husbands in treatment arms (34% vs 28%, p = 0.01; 72% vs 66%, p < 0.05; and 2.2 vs 2.0, p < 0.01, respectively) (see Table 1).

Adolescent wives were more likely to be lost to follow-up if they were nulliparous at baseline (15% nulliparous, 12% 1 birth, 8% 2 or more births; p = 0.02) or if their husband was polygamous (17% polygamous vs 12% monogamous; p = 0.06; data not shown). There were no differences in female retention rates across other demographics or across study arms.

The primary study outcome, reported current use of modern contraceptives among non-pregnant women, increased substantially over the study period, increasing from 11.8 to 38.3% overall; 17.0% to 29.2% among control participants and from 10.2 to 41.3% among intervention participants (p < 0.01; see Table 2). This change differed by specific intervention arm: modern contraceptive use among non-pregnant women increased from 6.3 to 40.0% in Arm 1, 17.1% to 40.4% in Arm 2, and 8.0% to 43.5% in Arm 3.

Women participating in the RMA intervention were more than twice as likely to report modern contraceptive use at follow-up relative to those in the control arm (adjusted IRR [aIRR] 2.33, 95% CI 1.41–3.87, p = 0.001; see Table 3). These intervention effects differed by intervention arm. In arm-specific analyses, women were significantly more likely to report modern contraceptive use at follow-up than control in Arms 1 and 3 (aIRR 3.65, 95% CI 1.51–8.78, p = 0.004 and aIRR 2.99, 95% CI 1.68–5.32, p < 0.001, respectively) (see Table 4). Arm 2 participants did not have significantly different likelihood of modern contraceptive use relative to control participants.

Post-hoc analyses to examine whether intervention effects on contraceptive use were limited to younger or older participants were consistent with overall models, showing significant associations between intervention and modern contraceptive use in both baseline age groups for Arm 1 (Age 13–16: aIRR 5.57, 95% CI 1.47–21.05, p = 0.01; Age 17–19: aIRR 3.14, 95% CI 1.21–8.15, p = 0.02) and Arm 3 (Age 13–16: aIRR 4.97, 95% CI 1.31–18.90, p = 0.02; Age 17–19: aIRR 2.56, 95% CI 1.34–4.89, p = 0.004) (see Additional file 1: Table S1). No significant association with modern contraceptive use was found for Arm 2 for either age group.

Post-hoc baseline parity-stratified analyses found no significant association between intervention and modern contraceptive use for nulliparous or multiparous women. However, women with one birth at baseline in Arms 1 and 3 had significant increases in modern contraceptive use (aIRR 6.67, 95% CI 1.54–28.85, p = 0.01; aIRR 5.57, 95% CI 1.83–16.96, p = 0.002, respectively) (see Additional file 1: Table S1). Arm 2 participants reported no significant associations with modern contraceptive use irrespective of parity.

Modern contraceptive use findings were robust to IPC weight sensitivity analyses, with direction and strength of associations for the time-by-treatment effects similar to the main regression models (Arm 1 aIRR 4.50, 95% CI 1.71–11.87, p = 0.002; Arm 2 aIRR 1.09, 95% CI 0.54–2.21, p = 0.81; Arm 3 aIRR 2.44, 95% CI 1.20–4.97, p = 0.01) (see Additional file 1: Table S2).

In terms of the secondary outcome, 8.9% of all wives across study arms reported IPV within the past year at baseline, increasing slightly to 9.8% at follow-up (Table 2). This increase was concentrated in the control arm (6.9% at baseline to 11.7% at follow-up), with the intervention arm remaining relatively static (9.5% at baseline to 9.2% at follow-up). This pattern over time differed by specific intervention arm: past year IPV increased from 3.9 to 9.0% in Arm 1, decreased from 10.6 to 7.2% in Arm 2, and decreased from 14.6 to 11.3% in Arm 3.

Women participating in any intervention arm of RMA were slightly less likely to report past year IPV at follow-up relative to those in the control arm, though this difference was not statistically significantly (aIRR 0.57, 95% CI 0.29–1.13, p = 0.11) (see Table 3). In arm-specific analyses, however, adolescent wives in Arm 2 were significantly less likely than those in the control arm to report past year IPV at follow-up (aIRR 0.40, 95% CI 0.18–0.88, p = 0.02) (Table 4). There was a similar magnitude reduction in IPV for married adolescents in Arm 3, though the findings were marginally significant (aIRR 0.46, 95% CI 0.21–1.01, p = 0.052). In contrast, participants in Arm 1 did not report a significant difference in past 12-month IPV relative to controls (aIRR 1.39, 95% CI 0.49–3.95, p = 0.54).

Post-hoc, baseline age-stratified analyses found a significant association between intervention and past year IPV only for wives ages 13–16 years in Arm 2 (aIRR 0.27, 95% CI 0.08–0.88, p = 0.03) (see Additional file 1: Table S3). Although the direction and magnitude of effects on IPV in both Arm 1 and Arm 3 were similar to those seen in the non-stratified analyses, regardless of age (aIRRs 0.36–1.55), no other intervention effects were significant among wives aged 13–16. No significant associations between intervention and IPV were found among wives aged 17–19 years.

Post-hoc, baseline parity-stratified analyses found marginally significant negative associations between intervention and past year IPV for nulliparous women in Arm 2 (aIRR 0.23, 95% CI 0.05–1.05, p = 0.06), and women with 2 or more births at baseline in Arm 3 (aIRR 0.41, 95% CI 0.15–1.13, p = 0.09) (Additional file 1: Table S3). As with age, the direction and magnitude of effects on IPV in Arms 1 and 3 were similar to those seen in the non-stratified analyses; no other significant intervention effects were observed across intervention arm and parity.

Past year IPV findings were robust to IPC weight sensitivity analyses, with direction and strength of associations for the time-by-treatment effects similar to initial adjusted regression models (Arm 1 aIRR 1.63, 95% CI 0.53–4.97, p = 0.39; Arm2 aIRR 0.41, 95% CI 0.17–0.96, p = 0.04; Arm 3 aIRR 0.45, 95% CI 0.20–1.00, p = 0.049) (see Additional file 1: Table S4).

No adverse events were recorded in this study.