Effect of a communication intervention on alcohol use during pregnancy in post conflict Northern Uganda: a quasi experimental study

Abstract

Background

Alcohol use during pregnancy is a preventable risk factor for Fetal Alcohol Spectrum disorders. Psycho-social and educational interventions have been reported to enable women reduce alcohol intake levels during pregnancy and help improve some health outcomes of unhealthy alcohol use. We set out to assess the effect of a communication intervention on alcohol use during pregnancy in post conflict northern Uganda.

Methods

The study employed a quasi - experimental design to assess the effect of a community health worker led communication strategy on pregnant women’s knowledge, attitudes and various patterns of alcohol use using Difference in Difference(DiD). 420 respondents were recruited at baseline as at endline.

Results

The communication messages were significantly associated with reduced odds of binge drinking (P = 0.018; OR = 0.09; CI = 0.012–0.66). Also those who received the intervention were less likely to drink frequently (P = 0.80; OR = 0.75; 95%CI = 0.074–7.5) or be harmful alcohol users(P = 0.948). The intervention also positively influenced having fair (β =0.49;P = 0.217;RRR =1.63)or adequate knowledge(β = 0.89;P = 0.25;RRR = 2.44) and having positive(β = 0.37;RRR =1.44;P = 0.46) or fair attitude(β = 0.19;RRR = 1.21; P = 0.693) although not to a significant level.

Conclusions

The communication intervention affected some patterns of alcohol use among pregnant women and not others. Our results contribute to existing evidence that communication interventions are a promising approach in reduction of alcohol exposed pregnancies. Interventions aimed at promoting alcohol abstinence during pregnancy should be implemented alongside other strategies that address factors that influence pregnant women to drink to achieve maximum results.

Background

Prenatal alcohol use is an avoidable cause of birth defects and developmental disabilities termed Fetal Alcohol Spectrum Disorders (FASD) [1, 2]. There is no cure for FASD and it is irreversible but it can be completely prevented if women abstain from alcohol consumption while pregnant or when trying to conceive given that women often do not recognize they are pregnant until 4 to 6 weeks post conception [3, 4]. A safe level of alcohol exposure during pregnancy is not known for the fetus, and the evidence is inconclusive regarding the effects of low to moderate maternal alcohol consumption [5, 6]. Furthermore, data consistently show that although many women reduce their consumption or abstain from drinking alcohol during pregnancy, some do not [7, 8]. Thus many pregnancies remain alcohol exposed across the world.

Global estimates of alcohol use during pregnancy in the general population were recorded at 9.8% with large variability between countries [9]. The World Health Organization (WHO) surveys have classified Uganda among the leading alcohol consumers globally in the last two decades [10, 11]. Worse still, in Uganda maternal drinking during pregnancy in the general population was estimated at 20.5%, the highest in Africa [9]. In post conflict northern Uganda, a region scarred from a two decade civil war, 23.6% of women seeking Antenatal care services(ANC) reported alcohol use (any amount) during pregnancy while 29.3% of drinkers reported binge drinking [12]. The relatively high levels of drinking during pregnancy in the region have been attributed to poverty and the deeply engrained social - cultural norms that don’t only tolerate but also glamorize drinking [13].

Women who drink at high-risk levels during pregnancy have other contextual issues that need to be addressed and these include depression, abuse, homelessness and among others [14].Psychological and educational interventions have been reported to promote alcohol abstinence or reduction of alcohol intake during pregnancy [15]. More so, adequate epidemiologic evidence links reduced levels of alcohol consumption with a reduced risk for morbidity and mortality, providing indirect support that behavioral counseling/ educational interventions that reduce acute and sustained alcohol intake levels can help improve some health outcomes of unhealthy alcohol use [16].

Health and non - health professionals have an important role to play in the prevention of prenatal alcohol exposure. In California, community health workers were instrumental in implementing a brief intervention (10 to 15 minute) guided by a scripted manual in community settings with much success as women in the intervention group were 5 times more likely to report alcohol abstinence as compared to those in the control group [17]. More so women in the intervention group had higher birth weight, birth lengths and lower fetal mortality [17]. In Northern Cape South Africa, community health workers were credited for raising awareness on harms related to drinking during pregnancy and FASD levels were reportedly lower in post intervention areas [18]. But these studies are few and far between especially in Africa.

To bridge this gap, this study assessed the effect of a community health worker led communication strategy on knowledge, attitudes and the quantity and frequency of drinking during pregnancy in a post conflict environment.

Methods

Study area and design

The study employed a quasi - experimental study design. We used the design of non-equivalent groups [19,20,21].The study assessed the effect of a communication intervention that was implemented from February 2021 to March 2022. Prior to that, a baseline study had been conducted in the same area. Factors found to influence alcohol use during pregnancy during the baseline were useful in informing the intervention. Kitgum and Pader districts were selected as the control group and Gulu district as the intervention district. Gulu district was the epi-center of much of the fighting between the Lord’s Resistance Army LRA) and Uganda Peoples Defence Forces (UPDF) soldiers. Kitgum and Pader district also experienced socio economic disruptions and deaths resulting from the war. Agriculture is the main source of livelihood for people in the three districts.

The communication intervention

The main communication goal was to increase knowledge, risk perception and improved practices related to alcohol consumption during pregnancy. The communication objectives were guided by Thomas W Valente [22] on the Hierarchy of Communication effects that postulates that when 80% of the population is exposed to the message, 60% will understand the message, 40% will approve the message, 20% will intend to Act and 5% will Act.

Specific objectives included 1) a 50% increase in number of women who know at least 4 dangers of drinking during pregnancy as compared to baseline; 2) At least 20% increase in number of women who believe alcohol use during pregnancy is important/very important in increasing risks to mother and baby during pregnancy and 3) By the end of the intervention, there will be a 5% increase in number of women aged 15–49 who have abandoned/reduced frequency and quantity of alcohol consumption compared to baseline findings.

Baseline study results [12, 13] were presented in a meeting during a community dialogue with men, women (target group) and health workers. In these meetings, a participatory approach was used in which stakeholders validated the study results and priorities were ranked with regard to which messages should be delivered. Later concepts were developed; communication materials were drafted with the Ministry of Health - Health Promotion Department. Messages were about chemicals contained in the alcohol, dangers of drinking during pregnancy, myths and misconceptions around alcohol use during pregnancy, factors that influence women to drink when pregnant and how they can stop it. Visual images to be included were also discussed with the stakeholders. Communication campaigns have been reported to be effective in raising awareness about drinking during pregnancy and that visual images can get the attention of the public towards understanding the consequences [18, 23]. These materials were translated to Acholi, the local dialect in the region by the Acholi language board.

Materials were then pretested using focus groups discussions with Village Health Team (VHTs) in the intervention district guided by a member from the district health department. The main points in the messages were summarized and VHTs made suggestions regarding the interpretation of messages, the font size, appropriateness of the images used, accuracy of the translation and the messaging. Revisions were made and a final leaflet made and printed.

Messages targeted women in reproductive age group as the primary audience but spouses, mother in-laws and other family members who directly influence the pregnant women’s decisions were also reached with messages. The sample size for the intervention was all women in the reproductive age group in Gulu district who are estimated to be 23% of the total population in the district according to UBOS 2018 projections. The VHTs visited all households with women of reproductive age group in their villages raising awareness about the dangers of drinking during pregnancy at least three times. These were supervised by health assistants/health inspectors.

We worked with 2 to 4 Community Health Workers known as Village Health Teams in Uganda (VHTs) per village depending on the size of the village and number of households therein. Each village has an average of about 30 to 80 households. Community health workers and other non -medical professionals have been instrumental in teaching women about dangers of maternal drinking. In De Aar and Upington, two areas heavily affected by FASD in the Northern Cape Province of South Africa, Community health workers taught women about FAS [18].

These VHTs were identified by health workers and trained by staff from the district health department. They disseminated these messages in their own villages where they hail from. The main mode of communication was interpersonal where VHTs moved from house to house raising awareness about the dangers of drinking during pregnancy. They also passed on these messages through already constituted groups such as village meetings, clan meetings, church gatherings and market days among others. This was deemed appropriate as some scholars have noted that communication interventions can target different levels of communities through wide varieties of mechanisms delivered by different modes of channels [24].

The VHTs were directly supervised by health assistants and filled in form showing number of households visited, numbers of women in those households, gatherings at which messages were disseminated, issues raised. These were signed and stamped by the village local council one chairman. These VHTs and the health assistants were paid a modest stipend agreed upon by the district health department and the research team each time they moved out to disseminate the messages.

Monitoring the intervention

After the first phase of the intervention, monitoring was conducted by the first author and research assistants using a simple monitoring tool adopted from a social behaviour change communication manual [25]. During monitoring of the communication campaign, Gulu district(intervention area) was stratified into sub counties. Simple random sampling was used to select the sub-counties in which the monitoring was done using paper lots. At the second stage parishes within those sub-counties were randomly selected. The villages were listed on paper and selected using simple random sampling. At the third stage, in each of the strata, households were selected from the house hold lists provided by VHTS.

The required numbers of households were obtained randomly. The sampling unit was women of reproductive age group and other influential family members. The Lot Quality Assurance Sampling (LQAS) which is a low - cost random sampling approach to monitoring and evaluation was used. It has been used a lot to monitor health interventions [26]. We had a random sample of 19(It provides an acceptable level of error). 5 Sub-counties were randomly selected out of the 10 sub-counties in Gulu District. Out of these 15 parishes were be randomly selected. We had a total sample size of 285 respondents. We asked them if they had received any messages about drinking during pregnancy in the communities and what messages they recalled from that interaction. This mini survey informed the next phase of intervention. VHTs in the control group were facilitated once to visit households reminding pregnant women about antenatal care visits at health facilities.

Sample size determination and sampling

With regard to the sample size for the endline we calculated the sample size by syntax; cluster sampsi, binomial detectable difference p1(.44) k(40) rho(.14) m [5].

Cluster-sampsi was used because treatment was assigned at cluster level yet outcome was measured at individual level.

P1 = Proportion in Population 1(control group).

P2 = Proportion in population 2(Intervention group).

K = Number of clusters in each arm.

Rho = Intra cluster correlation.

M = Sample size per cluster.

Power calculations

We estimated that a sample of 420 women (210 from control and 210 from intervention) from the two per cluster arm was needed. A health facility was taken to represent a cluster. Five women were randomly selected per health facility. This implies a random selection of 42 health facilities from each arm that provided the adequate sample size of 420 women. This allowed us to detect 14–17 percentage improvement in women abandoning the consumption of alcohol during pregnancy. Power calculations were done using stata version 15 software that provided us 80% power at 95% significance.

We used multistage sampling whereby we stratified Acholi sub-region into 8 regions consisting of the 8 districts. We then used simple random sampling to select the three districts of Pader, Gulu and Kitgum from the 8 districts in Acholi sub-region using paper lots. At the second stage, the health facilities providing ANC services in the three districts were randomly selected. All categories of health facilities in a given stratum were catered for purposes of representativeness. Health facilities were grouped first according to district, then later, according to levels (hospital, Health centre IVs, Health centre IIIs and health centre IIs). After that, we obtained the sampling interval by dividing total number of health facilities in sampling frame divided by facilities to be included in the sample n = 84. To pick the first facility, we rounded off the sampling interval up to highest full number and selected any number from that. We then obtained a random number between 1 and 2 that was chosen using paper lots. To obtain the next facility we rounded up the sampling interval to the previous result. We continued with this process until all required number of facilities were selected.

At the third stage, in each of the strata, the required number of respondents by health facility category were obtained randomly from the available facilities of the category of interest using systematic random sampling with the starting point obtained by simple ballot. The sampling unit was women attending ANC services at health facilities in the three districts where we obtained 5 women within each cluster. Unlike at baseline where sampling was based on probability proportional to size at endline sampling was based on power calculations to ensure the sample was powered enough to derive the required outcomes from the study.

Data collection

We interviewed women attending antenatal care at selected health facilities in Pader, Gulu and Kitgum districts in Post Conflict Northern Uganda. We assessed respondents for eligibility to participate in the interview. To be eligible, women had to be in the reproductive age group (at least 15 year and above) and be at least be (8 to 36 weeks) gestation and Ability to speak Luo and or English and be willing to participate. Women who refused to participate and women who are drank/very sick or required emergency care were not interviewed. Having sought ANC services, selected respondents were interviewed. We collected data using a computerized, Internet-based survey tool (ODK) using tablets. Women were asked about alcohol use (any amount and type). Quantity and frequency of alcohol consumed was also recorded. We also used the World Health Organization’ Alcohol Use Disorders Identification Test (AUDIT) questionnaire to assess alcohol related problems [27].

Amount consumed was quantified in terms of standard drinks WHO WHS(World Health Survey) defines a standard drink as one containing at least 8 to 13 g of pure alcohol [28]. The following measures were taken as one standard drink. i) a 285-ml bottle or can of beer, (ii) a 120-ml glass of wine (factory distilled or locally brewed), and (iii) a 30-ml glass/tot of a spirit or gin (factory distilled or locally brewed). For local brews a standard drink was measured using Alcohol by Volume for beverages where ABV was known [28]. The questionnaires were pretested in Amuru district. The questions in the AUDIT questionnaire were modified to reflect alcohol consumption during pregnancy.

We interviewed women about their Socio demographic, socio behavioural and other characteristics. In addition, we asked women about their knowledge of dangers regarding drinking during pregnancy and perceptions about the same. Women who reported 4 or more dangers of drinking during pregnancy were categorised as having adequate knowledge. Those who mentioned 2 to 3 effects were categorised as having fair knowledge and those who were able to mention 0–1 were regarded as having poor knowledge.

Data management and analysis

Data management and analysis was carried out using SPSS software program for windows version 21 SPSS Inc. Chicago USA as well as stata 15. Data entry errors were minimized by having a customized check program inbuilt in the data entry program and constant supervision by the researchers. Data was also checked for completeness and accuracy before leaving the facility and as such no data were missing. Frequency tables were generated for key variables in line with the analysis plan. Other forms of descriptive analysis included testing for association (e.g using chi-square test or t-test where applicable) between outcome variables as proxy measures of “alcohol related behaviour” and testing for significance change between intervention (Gulu district) and control (Kitgum and Pader). Such proxy measures of “behaviour change” included but were not limited to: knowledge, attitudes, frequency and quantity of alcohol consumed.

Difference-in-Difference (DiD) was used to measure the impact from the proxy measures of “reduced alcohol consumption”. For instance, where frequency of alcohol use was used to measure behaviour change then we computed the baseline value for both the intervention (Gulu) and control area (Pader and Kitgum) among the respondents. For instance let M₁ and M₀ represent frequency of alcohol use score for (Gulu) and (Pader and Kitgum) at baseline respectively while H₁ and H₀ are scores at endline. Then the DiD computation [29] was subjected to 5% level of significance test is represented as:

$$\textrm{DiD}=\left({\textrm{H}}_{1-}{\textrm{M}}_1\right)-\left({\textrm{H}}_{0-}{\textrm{M}}_0\right)$$

(1)

DiD = Difference in Difference.

H_{1 =} Observation for intervention group at endline.

M_{1 =} Observation for intervention group at baseline.

H₀ = Observation for Control group at endline.

M_{0 =} Observation for control group at baseline.

Multivariable analysis (DiD approach)

The DiD was preferred because it was not possible to randomize the intervention area and control area. We estimated the intention-to-treat (ITT) effect of the intervention, which captures the impact of being part of each group as opposed to actually receiving treatment. The ITT provides the average impact of the intervention on the population in general targeted by the program. The regression model (Eq. 2) controlled for key socio-economic and demographic characteristics e.g. sex, age, residence of respondents. The main statistical tests used included binary logistic, multinomial and OLS regression. Binary logistic regression was used in determining the effect of the intervention on alcohol dependence(non-alcoholic dependant or alcoholic dependent), alcohol use any amount (no or yes) frequent drinking(drinking at least three times a week i.e. frequent and infrequent), Binge drinking (light or Binge drinker).

Ordinary Least Squares Regression (OLS) was used to estimate effect of the intervention on harmful and hazardous drinking (The dependant variable was continuous/numeric).

The Multinomial Logistic Regression was used in determining the effect of the intervention on knowledge (poor, fair and adequate) and attitudes(poor, fair and positive)

$${Y}_{it v}={\sigma}_0+\kern0.5em {Treat}_{it v}{\sigma}_1+\kern0.5em {Post}_t{\sigma}_2+{ Treat XPost}_{it v}{\sigma}_3+\kern0.5em {X}_{it}\beta +\kern0.5em {\upsilon}_v+{u}_{it v};$$

(2)

Regarding the Audit, the hazardous alcohol use (questions 1 to 3) cut-off point was put at 6. The dependence (questions 4 to 6) cut-off was 4 and the harmful use (questions 7 to 10) cutoff was 7 points. Scores ≥8 were considered an indicator of problematic alcohol use [30].

Study participants were categorized as abstainers if they did not report alcohol consumption since on-set of pregnancy. Forms of alcohol consumption were 1) alcohol use (any amount); 2) Frequent drinking(at least three times a week); 3) Binge drinking (Consumption of four drinks at one sitting).

Results

Characteristics of study participants

Overall, 420 eligible women attending ANC in selected health facilities in Gulu, Kitgum and Pader districts in Northern Uganda were interviewed at end line as at baseline. Socio demographic characteristics of respondents at endline are described herein.

About a third of the respondents 132(31.4%) belonged to the 21 to 25 age-group. Most respondents 244(58.1%) reported Primary education as their highest level of academic attainment. The vast majority 374(89.05%) were either married or cohabiting and with regard to place of residence a significant proportion 255(60.7%) of the respondents hailed from rural areas. Slightly more than half 237(56.4%) belonged to the catholic religion and peasant farmers dominated the sample 114(27.1%) followed by house wives n = 109(25.9%). This is presented in Table 1.

Table 1 Socio Demographic Characteristics of Respondents at Baseline and End-Line

Other socio - behavioural and obstetric characteristics of respondents were captured; many of the women reported one sexual partner. Very few women reported to have ever smoked or used tobacco(1.0%) and less than 1% reported to have ever used marijuana. Most respondents at end-line were pregnant for the first time in both the control 73(34.8%) and intervention 68(32.4%) and majority were in their second trimester as at baseline. This is presented in Table 2.

Table 2 Social Behavioral and Obstetric Characteristics of Respondents at Baseline and End-Line

Association between the intervention and drinking patterns, knowledge and attitudes

Participants were asked at end line and at baseline about various patterns of alcohol use during the current pregnancy. Cross tabulation results reveal that the intervention was associated with reduction in alcohol consumption. The most notable impact was recorded with regard to binge drinking (consumption of four or more drinks at one sitting), where binging significantly (P = 0.04) reduced by 38.3 percentage points. Communication messages were also significantly (P < 0.001) associated with reduced consumption of alcohol (any amount) by 6.22 percentage points. However, there was a non-significant (P = 0.17) increase in frequent drinking by 6.55 percentage points but this was not attributed to the intervention. These are presented in Table 3.

Table 3 Impact of Intervention on Drinking Patterns (Cross Tabulation Results)

Regarding the AUDIT C scores, cross tabulation results show a decline in number of respondents reporting alcohol dependence by an insignificant (P = 0.97) proportion of 1.3 percentage point difference and only a 0.46 percentage point increase in participants insignificantly (P = 0.50)reporting hazardous drinking albeit these were not associated with the communication messages. These are presented in Table 3.

Respondents were asked a number of questions to gauge their knowledge concerning maternal alcohol use during pregnancy; Following the intervention, there was an 11.98 percentage point rise in women with adequate knowledge (women who could mention four or more danger of drinking during pregnancy) and this increase was highly (P = 0.001) significant. This is presented in Table 4.

Table 4 Impact of the Intervention on Respondents Knowledge and Attitudes(Crosstabulations)

A number of other proxy measures of knowledge appear to have improved and were significantly associated with the communication messages. These include; a 8.60 percentage point increase in number of women in agreement that pregnant women should not drink at all(P = 0.014); a 12.43 percentage point decline in women saying alcohol can be used occasionally during pregnancy(P = 0.001); a 10.64 percentage point decline in women mentioning that alcohol use is okay after attaining a certain number of months of pregnancy(P = 0.001); a 6.90 percentage point decline in women who noted that alcohol has no effect on the baby P = 0.040) and a 14.98 percentage point decline in women stating that alcohol can be good for pregnancy(P < 0.001). This is presented in Table 4.

With regard to attitudes, there was a7.46 percentage point increase in number of women with positive attitudes (felt alcohol was important/very important increasing health risks to the mother and baby during pregnancy) and this was significantly (P < 0.001) attributed to the communication intervention. This is presented in Table 4.

Effect of the intervention on drinking patterns, knowledge and attitudes

Binary logistic regression results reveal that women who received the intervention had reduced odds of indulging in binge drinking (P = 0.015; OR = 0.16; CI =0.038–0.707). More so, with the introduction of the intervention, alcohol use (any amount) and frequent drinking (drinking 3 times a week) decreased (P = 0.395; OR = 0.71; CI = 0.32–1.57) and (P = 0.85; OR = 0.84; CI = 0.14–5.04) respectively although not to a significant level. These are presented in table 5.

Table 5 Effect of Intervention on Drinking Patterns (Alcohol Use (Any Amount), Binge Drinking and Frequent Drinking)Binary Logistic Regression

Concerning the AUDIT scores, communication messages reduced chances of being a dependant drinker but not to a significant extent (P = 0.794; OR = 0.704;95% CI: 0.05–9.82). In the results of OLS regression, there was an AVERAGE decline in being a harmful alcohol user depicted by the negative coefficient of − 0.181 and being a hazardous alcohol user (β = − 0.172) although not to a significant extent (P > 0.05). This is presented in Table 6 and 7.

Table 6 Effect of Intervention on Audit Scores (Alcohol Dependence, Hazardous Use And Harmful Use) Binary Logistic Regression And OLS Regression

Table 7 Effect of Intervention on Respondents Knowledge and Attitudes (Multi-Nomial Logistic Regression

With regard to knowledge, a multinomial logistic regression was employed to predict the effect of the intervention on the probability of falling in any of the three knowledge categories; Poor knowledge, fair knowledge and adequate knowledge. The intervention had a positive influence in increasing knowledge from poor to either fair (β = 0.438, RRR = 1.55, P > 0.05) or Adequate knowledge (β = 0.968, RRR = 2.632, P > 0.05). However, the intervention had no significant influence in differentiating women with Fair or adequate knowledge relative to their counterpart with poor knowledge. This is presented in Table 8.

Table 8 Effect of the Intervention on Various Patterns of drinking(Multivariate Analysis)

We also used a multinomial logistic regression to predict the effect of the intervention on the probability of falling in any of the three attitude categories; Negative attitude, fair attitude and positive attitude. Results reveal that the intervention had a positive influence of enhancing attitudes from poor to fair (β = 0.090, RRR = 1.09, P = 0.833) or positive (β =0.363,RRR = 1.44,P = 0.25) but it had no significant influence in differentiating women with fair or positive attitude relative to their counterparts with poor attitude. This is presented in Table 6 and 7.

Multivariable results

After controlling for a host of socio-economic and demographic factors, the intervention remained significantly associated with reduced chances of binge drinking (P = 0.018; OR = 0.09; CI = 0.012–0.66). Also those who received the intervention were less likely to drink frequently although this was not significantly linked to the intervention (P = 0.80 OR = 0.75; 95%CI = 0.074–7.5).Surprisingly, the intervention was not linked to reduced odds of alcohol use(any amount)(P = 0.447; OR = 1.48;95%CI =0.53–4.10). This is presented in Table 8.

Concerning the AUDIT scores, results show that respondents who received the communication messages had reduced odds of being harmful alcohol users (P = 0.948). Receiving the communication messages was not associated with alcohol dependence (P = 0.384; OR = 7.08; 95%CI = 0.086–578.1 and hazardous drinking (P = 0.425; OR = 2.61;95%CI = 0.24–27.68). This is presented in Table 8.

After controlling for a host of other factors, multinomial logistic regression results reveal that the intervention positively influenced having fair (β =0.49;P = 0.217;RRR =1.63;)or adequate knowledge(Could mention at least 4 dangers of drinking during pregnancy) (β = 0.89;P = 0.25;RRR = 2.44) However it was not significant in differentiating women with fair or adequate knowledge relative to those poor knowledge. This is presented in Table 9.

Table 9 Effect of the Intervention on Knowledge (Poor Knowledge is the Base Outcome)

Similarly, results show that having received the messages on dangers of drinking during pregnancy was positively linked to having fair(felt alcohol was moderately important increasing health risks to the mother and baby during pregnancy) (β = 0.19;RRR = 1.21; P = 0.693) or positive attitude(felt alcohol was important/very important increasing health risks to the mother and baby during pregnancy)(β = 0.37;RRR =1.44;P = 0.46). None the less, the intervention was not significant in differentiating respondents with fair or positive attitude relative to those with poor attitude. This is presented in Table 10.

Table 10 Effect of Intervention on Attitudes (Negative Attitude is the Base Outcome)

Discussion

Results of this study show that the communication intervention on the dangers of alcohol use during pregnancy was significantly associated with reduced odds of binge drinking, a drinking pattern linked to Alcohol-exposed pregnancies (AEP), the direct cause of fetal alcohol spectrum disorders (FASD) [31,32,33]. This is in agreement with previous studies conducted in the region. A Community health worker led communication strategy in the Northern Cape Province of South Africa resulted in reduction in FASD rates and significant increase in knowledge on harmful effects of alcohol [18].

More so, in this interventional study on effects of a communication intervention on drinking during pregnancy, we observed that women exposed to the intervention were less likely to drink frequently and had reduced odds of being hazardous or harmful drinkers although not to a significant level. This is not dissimilar to previous studies employing psychosocial/ education interventions to reduce alcohol harms during pregnancy. Related studies have noted that promotion of alcohol abstinence alone had limited potential to impact behavior and should be done alongside other strategies meant to avert unwanted/unplanned pregnancies [34]. The other explanation for the non significant differences in hazardous, harmful and frequent drinking between the intervention and control arms may be attributed to the small numbers of women reporting these patterns of alcohol use.

It was also surprising that VHT home visits did not affect alcohol use (any amount)demonstrated by non-significant statistical tests between the control and intervention arms especially after putting into account a host of other factors. A similar study conducted in Southern California registered significant decline in alcohol use among those who received a 10–15 minute brief intervention administered by non - medical professionals as compared to those in the control group [17]. A qualitative inquiry might be helpful in explaining this phenomenon.

After controlling for other factors, the communication intervention positively impacted knowledge and attitudes towards maternal alcohol use during pregnancy but not to a significant extent.

This study is important because it is one of the few studies that have attempted to reduce alcohol consumption among pregnant women in Sub Saharan Africa and increase our understanding of alcohol use interventions in a post conflict setting. These results need to be interpreted in the context of the following limitations. We used self-reported measures of alcohol consumption that may be prone to recall and social desirability response biases. However, we tried to mitigate this by comparing with data from other sources or previous related studies. The best way to validate the self-report would be to use alcohol biomarkers.

Also, the study used the Difference in Difference (DiD) method to measure impact of the interventions and thus may not have easily estimated the role played by other prevailing factors that may affect differences in trends. Nonetheless, a multivariable linear regression was used to try and take into account a host of other factors. Both volume and concentration of alcohol consumed may be especially difficult to estimate where homebrews are common introducing bias. We used WHO World Health Service tool to define a standard alcoholic drink as one that contains as containing between 8 and 13 g of ethanol for commercial beverages whose alcohol content is known. The women interviewed at the baseline were different from the ones interviewed at endline as the intervention was evaluated after 12 months but all women in the reproductive age group in the intervention district were exposed to the communication messages and they bore similar characteristics as the women in the baseline. We did a simple random sample from a larger population and not a series of nested samples, a sampling design that could have created nonzero intraclass correlations within strata. As a result, the data may not be strictly independent.

Conclusion

The aim of this study was to evaluate the effect of a community health worker led communication intervention on alcohol use during pregnancy. Results of this study confirm that we can reduce the volume of alcohol consumed by women through raising awareness on its dangers. The communication messages on the dangers of alcohol during pregnancy affected some forms of alcohol use among pregnant women and not others. The intervention accomplished to some extent the communication objectives such as 6.22 percentage point decline in women drinking alcohol (any amount) and the 38.3 percentage point decline in women reporting binge drinking against the 5% set target. With regard to attitudes, the intervention wasn’t able to realise the set target of 20% as there was only a 7.46 percentage point increase in number of women with positive attitudes (felt alcohol was important/very important increasing health risks to the mother and baby during pregnancy).Also, the intervention fell short of realising the 60% target of women understanding the message as there was only a 11.98 percentage point rise in women with adequate knowledge. Our results contribute to existing evidence that communication interventions are a promising approach in reduction of alcohol exposed pregnancies. Interventions aimed at promoting alcohol abstinence during pregnancy should be implemented alongside other strategies that address other factors that influence women to drink alcohol to achieve maximum results.