An unmatched case-control study was conducted in nine hospitals in eight out of 25 districts of Sri Lanka over a period of six months. Five of these hospitals were selected for the study based on the highest frequency of all types of abortions reported in the Indoor Morbidity and Mortality Registers for each district [Medical Statistics Unit: Indoor Morbidity & Mortality Registers, unpublished]. Two hospitals were purposively selected to ensure adequate representation of the minority Muslim and estate sector Tamil populations. In Colombo district, both apex referral tertiary hospitals located in the commercial capital were included.
‘Cases’ were women admitted to the selected hospitals following an unsafe abortion. The potential cases were identified by consecutive screening of women admitted to the gynaecology and medical/surgical casualty wards with signs and symptoms suggestive of an abortion. Of them, women with a confirmed diagnosis of ‘induced abortion’ were identified based on the World Health Organization (WHO) criteria  and triangulated during in-depth interviews with each woman, under three categories: ‘certainly’, ‘probably’ and ‘possibly’ induced abortions. Details on the definition and sampling used to identify unsafe abortions are published elsewhere .
The ‘unmatched controls’ were postpartum mothers admitted to postnatal wards following the delivery of a term unintended pregnancy, which was defined by the pregnancy of a woman contracepting during the cycle of conception or not contracepting due to reasons other than desired pregnancy . They were selected from the same hospitals using a systematic random sampling method (every fifth mother in the postnatal ward admission registers) during the same study period.
The study aimed at recruiting a minimum sample of 159 cases and 600 controls, based on 80% power to detect potential associations between cases and controls at 5% alpha error; 20% minimum probability of exposure in the controls; with an odds ratio (OR) of 2.0; and 1:4 unmatched case-control ratio.
Using an interviewer-administered-questionnaire (Additional file 1), data were collected from the participants at an exit interview. The interviews were specifically conducted for this study as a separate research arm of a study that addressed several research questions on abortion. Pre-intern medical graduates who were not involved in providing care in the ward collected the data, after being trained in recruiting and collecting data from women by a group of psychologists and experts in qualitative research.
Data obtained from all cases and controls included: socio-economic status (age, marital status, education, employment), reproductive characteristics (gravidity, number of living male and female children), fertility behaviour (decisions made with partners on spacing and family size, family completion), reasons for never-use or discontinuation of contraception, and their contraception behaviour assessed in relation to three time points: ever, during last birth interval and during the cycle of conception. Among the ever-users of contraception, method that had been last used was also assessed. Modern methods were products or medical procedures that interfere with reproduction from acts of sexual intercourse, while traditional methods were natural/biological methods of interference.
Women were categorised by their contraceptive practices as ‘Ineffective users’ and ‘Non-users’ of contraception. ‘Non-users’ were further categorised as ‘Never-users’ (women who had never used contraception), ‘Early discontinuers’ (women who had discontinued contraception before the last birth interval) and ‘Late discontinuers’ (women who had discontinued contraception during the last birth interval before the cycle of conception).
Ethics clearance was obtained from the Faculty of Medicine, University of Colombo. This ethics approval was submitted to the provincial directors of health services and directors of each of the participating hospitals, upon which permission and ethics approval were granted to conduct the study. Informed verbal consent was obtained from each participant. Since induced abortion is illegal in Sri Lanka, obtaining written consent would dissuade patients from participating in the study, and therefore informed verbal consent was chosen over written consent. Participation was voluntary and participants could withdraw at any time.
Data were analysed using SPSS (Statistical Package for Social Sciences) version 20.
To assess the risk of abortion associated with individual factors (contraceptive practices, methods used, socio-economic and reproductive characteristics, fertility behaviour), all cases and controls were compared using crude odds ratio (OR) and 95% confidence interval (CI) in univariate analysis.
Next, to assess the risk of abortion associated with contraceptive practices after controlling for confounders, logistic regression analysis was performed to obtain adjusted OR. In the regression model (Model 1), case-control status was included as the dependent variable; and the main predictor (contraceptive practices categorised as ‘non-use’ and ‘ineffective use’) and confounders (all the factors significant in univariate analysis) as independent variables. The model was tested using backward likelihood ratio method at 0.05 probability of entry and 0.1 exit. Thereafter, the regression analysis was repeated (Model 2), with contraceptive practices further categorised as ‘never-use’, ‘early discontinuation’, ‘late discontinuation’ and ‘ineffective use’ to assess how the abortion risk would vary according to the long-term behaviour of cases and controls.
Finally, to assess how the type of contraceptive last used (traditional and modern methods) would modify the abortion risk associated with each contraceptive practice, three more regression models were developed among the ever-users of contraception: first with only the main predictor (type of contraceptive) and confounders as independent variables (Model 3), second with the addition of contraceptive practices (‘early discontinuation’, ‘late discontinuation’ and ‘ineffective use’) (Model 4), and third with the addition of an interaction term between contraceptive practice and type of contraceptive (Model 5). ‘Traditional method last used’ and ‘ineffective contraceptive practice’ were taken as the reference categories. If an interaction became significant in the model, to determine its joint effect with the respective contraceptive practice, odds ratio relative to the effect within the reference categories was calculated, by obtaining the exponential value (EXP) of the two beta coefficients of the contraceptive practice and its interaction.
Cases and controls were compared by the type of contraceptive last used among the ever users of contraception and reasons given for never-use/discontinuation among those who discontinued and never-users. Chi-square and Fisher tests were applied when assessing significance.