Effect of a community-based intervention to increase participation in cervical cancer screening among Pakistani and Somali women in Norway

This study was originally designed as a community-based cluster randomised intervention trial in four municipalities surrounding the capital Oslo (Lørenskog, Bærum, Asker and Drammen). Geographic areas within the municipalities were matched in pairs according to the number of Somali and Pakistani women living there and randomized as clusters. However, at the beginning of the intervention period we observed the presence of tight social bonds and common meeting places between the immigrants across the areas we had depicted. Immigrant women in the intervention group had family and friends in the control areas and sometimes women living in the control areas showed up to participate in meetings in the intervention areas. We realized that, in practice, discussion among the women about the intervention in control areas was unavoidable, resulting in a large risk of contamination of the control group. For this reason, we abandoned the cluster-randomized design and instead compared women living in the areas randomized to the intervention group against women living in Oslo. Oslo was chosen as control group because the city is geographically close to the four municipalities included in the trial and has similar demographics.

We included all women with immigrant background from Somalia or Pakistan (born in Somalia or Pakistan or Norwegian-born with Somali or Pakistani parents) who lived in the intervention areas or the control area as per January 1st 2017, were between 25 and 69 years of age January 1st 2017 and were registered as living in Norway throughout the whole period between January 1st 2012 and January 1st 2018.

For the original randomized study, the necessary sample size was calculated for an expected increase in cancer screening participation from 45 to 55%, 80% power and 5% significance level. At the time this study was developed, there was no available data on the percentage of immigrant women who participated in the CCS programme in Norway, and based on international studies [20, 21] and our own proxies [10], we conservatively estimated 45% screening participation in the control group. We obtained information on the number of Pakistani and Somali women aged 20–69 who lived in the four target municipalities from Statistics Norway [10]. When applying an average number of women of 612 per cluster (geographical areas with substantial female immigrant population) and an intra class correlation (ICC) of 0.025 in the sample size calculation, 32 clusters (16 intervention clusters and 16 control clusters) were enough. The 32 clusters from four different municipalities were matched in pairs according to the number of female immigrants aged 20 to 69 from Somalia and Pakistan and randomized pairwise. All Somali and Pakistani women who satisfied the inclusion criteria and were living in one of the 16 clusters randomized to receive the intervention were included in the intervention group (n = 1554). Because of the contamination described earlier we did not include women living in the 16 control clusters in the analyses as controls, but instead chose to use all Somali and Pakistani women living in Oslo as the control group (n = 9266). If the proportion screened is 45% in the control group, the minimal detectable difference in proportion screened with 80% power and 5% significance level with this sample size is 3.8%.

The principal author, a senior researcher of Pakistani origin, approached the Pakistani women by mobilizing different channels such as organizations and community centers to identify one informant in each of the four areas. With the use of personal social circle and involvement of organization and community centers, including the imams of the mosques in the intervention areas, four key informants were recruited, and they further recruited participants to a meeting in the area where they lived through different channels such as personal phone calls, community gatherings and social media (Facebook). Women who did not show up for the first meeting despite having agreed to it, were contacted again to attend a second meeting through personal visits, phone calls and through women who attended the first meeting. A Somali female research assistant recruited the Somali women adopting similar procedures. Among the 1123 Pakistani and 431 Somali women who were living in the intervention areas as indicated by The Statistics Norway (SSB) the key informants managed to invite a total of 401(36% of the Pakistani) and 378 (88% of Somali) women to one of the seven meetings arranged for each group. Among them, 102 (9% of the total and 25% of invited) Pakistani and 128 (30% of the total and 34% of invited) Somali women attended the meetings.

The intervention was carried out from February to October 2017. The details of the intervention are already published [19]. It consisted of a power-point presentation with a general description of our project aims followed by brief information on healthy lifestyle and preventative health care and then narrowing it down to cervical cancer. In very simple language (Somali or Urdu), we explained what cervical cancer is, its anatomical location -through diagrammatic illustration-, causes, risk factors, and development. This was followed by practical information on cervical cancer screening including the procedure and the instruments used, which were also shown physically. A short video clip of the test was played. In addition, the women were informed about the Norwegian screening guidelines for cervical cancer and on how to book an appointment with their GP for taking a screening test, including information about the payment. The content of the presentation (power point) was both in English and in the respective languages (Urdu and Somali).

For the evaluation of the intervention, we obtained data through three national registries, linked by the unique personal identification number available for each Norwegian resident.

Information on screening status at baseline and post-intervention was obtained from the Norwegian Cancer Registry where women between 25 and 69 years old are registered when they take the CCS test, provided they consent to registration. The main outcome measure was screening status as per 1st of January 2018. We defined a woman as screened if she had taken a CCS test within 3 years before January 1st, 2018. Likewise, a woman was defined as screened at baseline if she had taken a CCS test within 3 years before January 1st, 2017. A three-year time window was chosen since the recommendation in Norway is to take a CCS test every third year.

Information about age per January 1st, 2017, years since immigration, country of origin, country of birth, marital status, income in 2016, and highest achieved level of education was obtained from Statistics Norway. Income was divided into quartiles based on the distribution of income for all Pakistani and Somali women aged 20–69 in Norway in 2016.

Information on the woman’s GP including GP’s age, gender and country of origin was obtained through linkage of data from the national GP database and Statistics Norway.

Baseline characteristics for the intervention group and the control group (Oslo) were compared using descriptive statistics with means and standard deviations for continuous variables and counts and percentages for categorical variables. Since the allocation of the intervention was not randomized between the two groups, we tested for baseline differences using t-tests for continuous variables and chi-square tests for categorical outcomes. We adjusted for education in a separate model as there were a lot of missing values for this variable. The possible reason could be that missing registration of education taken in the country of origin.

The effect of the intervention was estimated by applying generalized estimation equations (GEE) with binomial distribution and identity link function with screening status as the outcome using data in long format with two binary measurements of screening status per person (status January 1st 2017 and status January 1st 2018).

The effect of the intervention was obtained as the interaction term between a binary time-variable and the binary variable for intervention group in a model also containing a main effect for time and a main effect for intervention group. Effects on an absolute scale were obtained as the coefficient for the interaction term between time and intervention group and reported as risk differences (RD) with 95% confidence intervals. The confidence intervals were estimated with robust standard errors. The interaction effect can be interpreted as the difference in change in the absolute proportion screened with positive values indicating that screening participation has increased more in the intervention group compared to the control group.

We applied four levels of covariate adjustment; unadjusted (Model 1), adjustment for woman’s age, country of origin, first/second generation immigrant, marital status and income quartiles (Model 2), additional adjustment for GP’s age, GP’ gender and GP’ country of origin (Model 3) and finally additional adjustment for the woman’s educational level (Model 4). The adjustment variables were chosen because they have been found to be associated with participation in screening in previous studies [10, 11] and because they differed between the two groups at baseline.

As additional analysis we also applied propensity score stratification with deciles of propensity scores instead of traditional covariate adjustments. The propensity scores were obtained as predicted probabilities of being in the intervention group from a logistic regression model with intervention-group as the outcome and woman’s age, country of origin, marital status, income quartiles, GP’s age, GP’ gender and GP’ country of origin as independent variables.

As a second sensitivity analysis we applied the same GEE-models as described earlier to compare change in screening status from January 1st, 2016 to January 1st, 2017 between the intervention group and the control group. Since this was before the intervention, any differences would have to be caused by something else than the intervention.

We also did subgroup analyses to compare the strength of the effect of the intervention between Somali women and Pakistani women and between first and 2nd generation immigrants. Effect modification by country of origin and by generation was evaluated by including a 3-way interaction between intervention group, time and country of origin in GEE models.

All tests are two-sided, and 5% significance level was applied in all analyses. All analyses were conducted using Stata MP version 16.

This is the first of its kind study to evaluate the effect of a community-based intervention among immigrant women in Norway. One of the key strengths of this study was the measurement of cancer screening using registry data rather than self-report, which tends to overestimate screening and varies with ethnicity [35, 36].

The most important limitation was that we had to change the initial design of cluster randomization because of possible contamination. We are aware that, the lack of randomization poses challenges such as selection bias and confounding. Which not only questions the validity and usefulness of the findings, but in addition non-randomized design is associated with a high risk of bias because known and unknown characteristics may be distributed unequally between groups [37]. Same may be the case with our findings, we cannot rule out that the observed effect is at least partly caused by residual confounding. In addition, contamination between the two groups may have decreased the effect size. However, all analyses conducted to explore this possibility point to a real effect of the intervention. Previously published studies have highlighted the methodological challenges of recruiting minority groups into research trials [24‐27]. Another limitation worth mentioning is, we did not include men as targets of intervention in our study. Health behavior changes of an individual are complex processes where the individual, her surrounding family, health care providers and the health system mediators take part to achieve a positive outcome. In cultures where males have the power of decision in regarding the participation of women in the screening program it might have been positive to include the male members of the Somali and Pakistani communities and give them information on CCS and why it is necessary [32, 33].

Our study showed increased CCS after attendance to one short meeting with educational information, the effect of which was measured within 1 year after the intervention. Thus, we are unable to comment on the long-term effect of the intervention which may not be visible in a short time in fact the effect might not be sustainable without reminders. Unfortunately, our study material did not allow to have a longer study period.

Cost-effective analyses of both the interventions separately, the intervention among immigrant women and the intervention among GP done previously by the research team, as well as a study including both interventions simultaneously and over a longer time period, would help shaping and adaption of the right policy. Our findings also suggest that we need comprehensive measures directed at increasing participation in screening and preventing cervical cancer among immigrant women.