Measuring contraceptive self-efficacy in sub-Saharan Africa: development and validation of the CSESSA scale in Kenya and Nigeria☆☆☆

Objectives Contraceptive self-efficacy, a women's belief about her own ability to complete the actions necessary for successful family planning, is a well-documented determinant of contraceptive use. However, there is currently no validated measure appropriate for low-resource settings. We developed and tested a new scale to measure Contraceptive Self-Efficacy among women in sub-Saharan Africa (CSESSA) using samples in Kenya and Nigeria. Study design The CSESSA scale was administered to women in Kenya (n = 314) and Nigeria (n = 414). Reliability and validity were analyzed separately by setting. Validity analysis included assessment of the area under the curve (AUC) to demonstrate predictive capability of CSESSA score for contraceptive use. Logistic regression was employed to test the relationship between CSESSA score and contraceptive use. Results Item reduction resulted in 11 items in Kenya (α = 0.90) and 10 items in Nigeria (α = 0.93). Three domains of contraceptive self-efficacy emerged in both settings: (1) husband/partner communication, (2) provider communication and (3) choosing and managing a method. Items related to the first two subscales, but not the third, were identical across settings. The AUC indicated predictive capability as mild in Kenya (AUC = 0.58) and strong in Nigeria (AUC = 0.73). In both settings, CSESSA score was associated with use of a modern contraceptive method at 12 months postpartum. Conclusions The CSESSA scale is a reliable and valid measure in two countries. Variation of the third subscale by site indicates that certain scale items may be more relevant in areas of low versus high contraceptive prevalence. Further research should be done to validate this subscale in other contexts. Implications This study contributes a reliable, valid measure of contraceptive self-efficacy in two African countries. The CSESSA scale and subscales can be administered in research (for example for evaluation of interventions to increase contraceptive uptake) or in a clinical setting to inform and improve contraceptive counseling.


Introduction
Theories of self-efficacy link an individual's beliefs about their personal capabilities to their health behaviors, and evidence indicates that practices that promote self-efficacy influence behavior change [1]. Contraceptive self-efficacy (CSE) is a woman's belief in her own ability to succeed in contraceptive initiation, management and continued use. Despite being a recognized precursor to effective contraceptive uptake [2][3][4][5], CSE is not routinely measured in low-resource settings. Most efforts to measure contraceptive-related self-efficacy have been restricted to high-resource settings [2,3].
Levinson developed a scale to measure CSE among adolescents in high-income contexts which has been validated in a variety of settings across the United States, Canada and Mexico [2,3]. Most items in Levinson's scale are not relevant to nonadolescent women in lowresource settings as they pertain to adolescent-specific subjects such as parental knowledge of contraceptive use. There are validated scales to measure self-efficacy for condom use, sexual communication and protective sexual behaviors [6][7][8][9]. However, review of the literature revealed no appropriate standardized tool to measure CSE in low-resource settings such as sub-Saharan Africa (SSA).
A validated measure of CSE for a low-resource context would strengthen contraceptive research and could be used to develop, evaluate and improve contraceptive promotion efforts. Responses to items on a CSE scale could identify specific behaviors for which women have low

Contents lists available at ScienceDirect
Contraception: X j o u r n a l h o m e p a g e : h t t p s : / / w w w . j o u r n a l s . e l s e v i e r . c o m / c o n x self-efficacy that can be addressed by activities to promote contraceptive adoption and continuation. Exploring CSE levels over time could lend insight to factors that influence CSE and how CSE may mediate contraceptive behavior.
We developed a new scale aiming to measure Contraceptive Self-Efficacy among women in sub-Saharan Africa (CSESSA). This paper presents findings from reliability analysis and validation of the CSESSA scale in two independent samples of women in Kenya and Nigeria.

Scale development
We assessed the transferability of Levinson's CSE scale to the SSA context through conversations with co-investigators and reproductive health program staff in Kenya and Nigeria. Concepts underlying certain scale items, such as partner communication and seeking contraception from a health provider, were selected for inclusion based on relevance to the target population. We reviewed the literature on factors influencing contraceptive behavior in SSA and hypothesized that CSE comprised of four domains: husband/partner communication, friend/family influence, provider communication, and choosing and managing a method [10][11][12]. Findings from focus group discussions with women of reproductive age in Nigeria informed item development for these four domains. Bandura's theories on self-efficacy guided item phrasing. Iterations of scale items were reviewed and revised by the authors with in-country colleagues and remotely by a group of reproductive health experts. The resulting 21 items were pilot-tested among 8 postpartum women attending health clinics in Kenya. Revisions were made, and three double-barreled items were removed. Eighteen items remained for scale development (Appendix A).

Participants
The 18-item CSESSA scale was administered to women participating in a cluster randomized trial of group-based antenatal and postnatal care in Kisumu and Machakos counties, Kenya, and Nasarawa State, Nigeria. Kisumu is located in Western Kenya on Lake Victoria, and Machakos borders Nairobi. In both counties, nearly all pregnant women receive antenatal care from a skilled provider, and the modern contraceptive prevalence rate (mCPR) is higher than the national average (59% in Kisumu and 68% in Machakos compared to 53% nationally) [13]. In Nasarawa, a diverse, underresourced state in central Nigeria, 77.1% of pregnant women receive antenatal care from a skilled provider [14]. The mCPR is slightly higher in Nasarawa State compared to Nigeria overall (14.3% compared to 12.0% nationally) [14].
Inclusion criteria and study methods are detailed elsewhere [15]. The present analysis uses cross-sectional data from a survey administered at the end of the study to participants who were 12 months postpartum. To remove any potential effect of the intervention on CSE, data are constrained to participants in control facilities only.
Due to inclusion criteria of the parent study, the sample is further limited to women who attended at least one antenatal care visit before 24 weeks' gestation, consented to participate in the study and were available for follow-up 1 year after delivery. Women whose infants died before 12 months postpartum are excluded from analysis (n = 1 Kenya, 20 Nigeria) under the assumption that these women may desire another pregnancy or be pregnant at the time of survey (12 months after delivering). Data collection was completed in July 2018 in Kenya and March 2018 in Nigeria.

Procedures
Study staff contacted participants by phone or in-person to schedule the 12-month postpartum survey. The survey was then administered to participants in their homes by research assistants (RAs) using RedCap mobile technology to upload data remotely from a tablet to a secure server. All questions were read aloud to participants. For the CSESSA scale, the RAs asked women to rate the certainty with which they could do each item (for example, discuss family size with my husband/ partner). A visual analogue scale was used as an aide to describe the response options, which ranged from 0 (cannot do at all) to 10 (highly certain can do). As needed, RAs asked participants to clarify a response falling between two tick marks and recorded the value closest to the point on the line indicated by the participant.
The survey also collected information on participants' sociodemographic characteristics and contraceptive behaviors. An indicator of household wealth was generated based on methods used by the Demographic Health Survey [16]. Women were considered to be current modern contraceptive users if they responded positively to the question "Are you currently using a family planning method to prevent pregnancy?" and reported current use of condoms, oral contraceptive pills, injectables, implant, intrauterine device, emergency contraception or sterilization.

Sample description
The samples in both settings surpassed the size required by the ratio of subject-to-item guidelines of 10 subjects per item and demonstrated 100% response rates for each item [17]. Demographic characteristics of study participants are provided in Table 1. Differences are notable between settings. While most women in the Kenyan sample (73.6%) were Protestant and almost all (98.1%) were literate, women in the Nigerian sample were predominantly Muslim (73.9%) and just over half (56.8%) were literate. Most women in Kenya recently delivered their first or second child (59.9% vs. 46.4% in Nigeria), while over a third of the Nigerian sample recently delivered their fourth or fifth child (35.2% vs. 19.4% in Kenya). Modern contraceptive use at 12 months postpartum was high in Kenya (73.3%) and low in Nigeria (27.5%).

Ethical clearance
This study was reviewed and approved by the Johns Hopkins Bloomberg School of Public Health Institutional Review Board, the Kenya Medical Research Institute Ethics Review Committee and the National Human Research Ethics Committee of Nigeria. Written informed consent was obtained from all participants prior to data collection.

Analysis
We conducted psychometric analyses to assess the reliability, fit and structure of the 18-item scale independently in each setting. The reliability coefficients were high in both Kenya (Cronbach's α = 0.93) and Nigeria (α = 0.97). To remove redundant and poor-fitting scale items, we assessed item-test and item-rest correlations and factor loadings derived from principle components analysis (PCA). Items having itemrest or item-test correlation < 0.60 were removed [18]. Items were retained if their greatest factor loading was >0.60 and second highest was <0.30 [19]. Items with uniqueness above 0.50 were removed, leaving a total of 13 items remaining for further analysis across sites (retained items presented in Table 2). Of these 13 items, 8 were consistent across sites. Some items that were retained in Kenya were removed in Nigeria (n = 3) or kept in Nigeria but not Kenya (n = 2).
Internal consistency of the scale with 11 items in Kenya and 10 items in Nigeria was reassessed by Cronbach's α. PCA was then performed to determine the structure of the scale; three factors produced eigenvalues above 1 and were visible with a scree plot. Parallel analysis confirmed a three-factor composition [20]. These were extracted through exploratory factor analysis using promax rotation and retained as the following scale domains: husband/partner communication, provider communication, and choosing and managing a method.
Reliability analysis was conducted to assess the potential of subscales by domain. Items for husband/partner communication and provider communication were identical in both settings. However, items related to choosing and managing a method varied, signaling that this domain may manifest differently in areas with low and high contraceptive prevalence (see Appendix B). Mean scores for the full scale and for each potential subscale were calculated by dividing the summative score by the number of scale items, leading to a standard range of 0-10 points. Means were compared across age, education, parity and household wealth by t tests.
Validity of the scale was then assessed separately for both samples (see Appendix C). We calculated the area under the curve (AUC) of a receiver operating characteristic curve to assess the predictive capability of the total CSESSA score against current modern contraceptive use (criterion-related validity) [21,22]. To address imbalance in the data, we also calculated the AUC for the total score grouped into quartiles.
Construct validity was assessed through logistic regression of the total CSESSA score against current modern contraceptive use [21]. Generalized estimating equation was used to account for clustering of data at the health facility level (10 clusters per site) [23]. For sensitivity analysis, we also performed multivariate logistic regression of total scores grouped into quartiles as a predictor of current modern contraceptive use. All data were analyzed in Stata15.  Table 3.

Validity
As shown in Fig. 1, the AUC of 0.58 suggests that while a woman's total score on the CSESSA in Kenya may predict her willingness to use modern contraception, more investigation into this relationship is required. Grouping the total score into quartiles revealed an improved predictive capacity when the distribution of scores was more balanced (AUC = 0.61, standard error 0.03). Each of the three potential subscales produced a similar result with AUC ranging from 0.51 to 0.58 (Fig. 2).
Results of multivariate logistic regression indicate a valid measure: for each one-point increase in the CESSA total score, the odds of current modern contraceptive use increased by 4.0% [adjusted odds ratio (aOR) 1.04, p < .001, confidence interval (CI) 1.02-1.06] controlling for age, language, religion, education, parity and household wealth. This same relationship held when the total CSESSA score was split into quartiles (aOR 1.73, p < .001, CI 1.35-2.21), controlling for the same variables. We ran the same model to assess the relationships of each subscale to current modern contraceptive use. The total score from each subscale was significantly associated with the outcome as shown in Table 4.

Reliability
Assessment of internal consistency for the 10-item scale in Nigeria indicated strong reliability (α = 0.93, ICC 0.56). On a single-factor solution, the item Discuss specific family planning methods with my husband/ partner loaded highest.  provider communication, and 6.60 (SD 3.38) for choosing and managing a method (Table 3).

Validity
The AUC of 0.73 (standard error 0.03) indicates that a woman's total score on the CSESSA in Nigeria has a strong predictive capability for modern contraceptive use (Fig. 1). Grouping the total score into quartiles reduced the predictive value slightly to an AUC of 0.70 with standard error of 0.03. Similarly, all three subscales were highly predictive of current modern contraceptive use, with AUC above 0.67 (Fig. 2).
In the Nigeria sample, each one-point increase in CESSA total score (0-100) increased the odds of current modern contraceptive use by 6.0% (aOR 1.06, p < .001, CI 1.05-1.08) controlling for age, language, religion, education, parity and household wealth. In the same model, each progressive quartile of total CSESSA score increased the odds of current modern contraceptive use 3.02 times (aOR 3.02, p < .001, CI 2.23-4.09). In separate multivariate regressions, each subscale total score was statistically significantly associated with current modern contraceptive use (Table 4). *p < .05, **p < .01, ***p < .005. a t tests assessed differences in mean score by sociodemographic characteristic, with significant results indicating a relationship between the sociodemographic characteristic and mean score.

Discussion
This paper contributes a reliable, validated measure for CSE in two African countries. While the full scale provides a multifaceted and comprehensive measure of CSE, subscales allow researchers to focus on specific aspects of CSE using fewer items. The three scale domains align with factors identified in the literature as having influence on contraceptive use in SSA [10][11][12]. Notably, items pertaining to the fourth hypothesized domain of "friend/family influence" were not retained in statistical analyses, indicating that social influence may be less relevant to CSE in these settings. Further research is warranted, such as cognitive interviewing to assess the response process for these items. Establishing scale reliability and validity in two distinct settings is a strength of this study. However, given that the "choosing and managing a method" domain differed for the two settings (generating two versions of a similar subscale), each should be separately tested in other contexts with similar levels of contraceptive prevalence. For example, the three-item subscale which we tested in Nigeria should be assessed to see if it is reliable and valid in other low-mCPR settings, while the four-item subscale that we tested in Kenya should be similarly assessed in higher-mCPR settings.
In assessment of criterion-related validity, the AUC in Kenya for quartiles of total score fell just above our cutoff point (0.60), while the Area under the receiver operating characteristic curve AUC of continuous total score and the total of each subscale fell below.
These findings indicate either that current modern contraceptive use may not be an appropriate "gold standard" measure for CSE or that CSE may not serve in a predictive capacity for modern contraceptive use in this context. In contrast, validity results were consistently strong in Nigeria. Regression results show a strong relationship between CSESSA score and current modern contraceptive use in both settings. The CSESSA scale and subscales can be used in contraceptive research and programming to pinpoint key intervention opportunities and to evaluate program effectiveness. Assessing mean scores by subscale and demographics, as we demonstrate in Table 3, can help to identify those who need contraceptive care support. For example, our findings indicate that women with lower education in Nigeria may need more support to communicate with their husband/partner about contraception. Our findings related to influence of age [24][25][26], education [11,26] and wealth [25,26] on CSE are consistent with literature on these factors' influence on contraceptive uptake (Table 3). Mean scores were higher in the Kenyan as compared to the Nigerian sample, reflecting differences in norms of contraceptive use and prevalence between the two settings. Additional research should be done to assess how prior contraceptive use influences women's CSE and the measure's relation to contraceptive discontinuation.
The mCPR in both sample populations is higher than national estimates (Kenya: 73.3% vs. 53.0%; Nigeria: 27.5% vs. 12.0% nationally) [14,15]. Given the eligibility criteria for the parent study, these samples may represent a specific subset of women who have higher care-seeking behavior, likelihood of using contraception and CSE compared to women who were not recruited.
This study has several limitations. First is the fact that the scale was included in a postpartum survey resulting in a homogeneous sample in terms of variation in reproductive experiences, health system interactions, preferences and partner status. Items pertaining to husband/partner communication may be less relevant to women at various stages of life and in different settings. This homogeneity may contribute to high IICs, particularly in Nigeria. Also, the cross-sectional nature of administration is not ideal. Preferably, the scale would be administered at multiple time points to gauge whether levels of CSE fluctuate or hold constant relative to an intervention.
Lastly, in the Kenyan sample, distribution is heavily skewed toward high scores. This likely is a reflection of Kenya's high mCPR; however, it is also possible that social desirability bias or survey fatigue contributed to consistently high scores. The CSESSA scale appears to be a reliable, valid measure of CSE for women in both Kenya and Nigeria.