The study was conducted in one rural and one urban site in South Africa. The sites were selected because there was infrastructure in place to support the recruitment of patients. The urban site in Cape Town was primarily located at a regional hospital and the surrounding communities. The service for people who had been raped was established by the department of justice and is located at the hospital. There is a full time nurse who is primarily responsible for follow-up visits. Any available emergency department physician carries out examinations. The rural site, a district level hospital, was located in the far north of the country and serves the town of Thohoyandou and the surrounding rural area. A non-governmental organisation has established a trauma centre at the hospital to deal with sex based violence and provides support and counselling. Any doctor at the hospital carried out examinations.
We interviewed 319 women at the two sites between November 2003 and January 2004. In Cape Town we recruited 74 women who had used the service and four carers of patients who were too young or could not to be interviewed due to mental disability. In Thohoyandou we recruited 81 patients. We also interviewed 78 comparable women from the community in Cape Town and 82 in Thohoyandou. They were of similar age (plus or minus 5 years) and lived in the same area (living 10 doors away from each service user). During recruitment no connection was made between the community participant and the service user, in this way confidentiality was maintained. They were selected to determine whether potential service users (including those who may have chosen not to use the service) have different priorities from those who actually access services. Women who had survived were approached through the health services over the six months before the survey. The examining doctor or nurse in participating services asked them if they would consent to their details being given to a researcher studying quality of services. Those who agreed gave written consent. At the time of the survey, the women were asked again for consent to participate in the study. Though we did not exclude men who had been raped, none were recruited.
Trained fieldworkers administered the questionnaire through face-to-face interviews in Xhosa, Afrikaans, and Venda. We collected data on sociodemographic characteristics (including socioeconomic status17
), attitudes towards rape services, and discrete choice responses from all participants. Participants who had used health services were asked about the actual management that they received including referrals for counselling, treatment received, side effects experienced (if any), information received on side effects, and number of return visits.
The attributes and levels used in the discrete choice analysis were based on results from focus groups held in different parts of the country, expert inputs, and pertinent policy questions. The box outlines the attributes and their levels. This combination resulted in 1162 possible scenarios, which was reduced to 16 by using a main effects orthogonal fractional factorial design in the SPSS orthoplan procedure. We selected one scenario, with services closely reflecting current levels, as a constant comparator (service A) and then presented 15 choice scenarios between this and service B ( shows an example). We used a pictorial representation of the scenarios to facilitate selection by participants.
Sample of discrete choice analysis item from questionnaire. Participants had to choose service A or service B*
We included one scenario with all attributes of equal or inferior levels to check that study participants really understood the exercise and were making deliberate choices. We then excluded from any further analysis the 23 participants who did not select the logically better scenario. We excluded respondents who always selected choice A or B, as well as those who obviously failed a test of transitive preferences (which assumes that if bundle A is preferred to B and B is preferred to C then A should be preferred to C) (n = 21). Our final analysis included 275 participants (86% of original sample); 135 who had been raped (or a carer responding on behalf of a patient) and 140 community participants; 121 were from Cape Town and 154 were from Thohoyandou.
Attributes and levels in the discrete choice analysis
Travel time to service
- < 30 minutes
- 1-2 hours
- ≥ 3 hours
Availability of HIV prophylaxis
- Not available
- Available if the results of the HIV test were negative
- Available without HIV test being done
Required returns to the services
- No return after initial visit
- One return after three days
- Four returns, once a week for one month
- Short examination that will document physical injuries like bruises, cuts and grazes but the main focus will be on your health
- An extensive examination to collect evidence that may result in a greater likelihood of the case going to court
Skills and attitudes of nurses and doctors
- No understanding, no training in counselling
- Understanding but no formal training in counselling
- Understanding and formal training in counselling
Data were entered into SPSS 11 (SPSS, Chicago, IL) and analysed in Stata 8.0 (StataCorp, College Station, TX). We used bivariate analysis for social characteristics and health services received by participants. The two sites were compared with χ2 test for categorical variables and a double sided t test for continuous variables.
Because of the multiple response nature of the data we used a random effects probit model to estimate the benefit function of moving from service A to service B. A likelihood test of r =0 showed that this was the correct model to use (P < 0.001). We expressed attributes with ordinal or categorical levels as dummy variables (all but number of returns to the service required) as we could not assume linearity, with the benchmark being the level in the comparator scenario (see ). Models included variables for the attributes and levels as well as for potential interaction effects with socioeconomic and demographic characteristics and previous service use.
We used likelihood ratio tests to determine coefficients on interaction terms with site, service user, socioeconomic status, and age. Site interactions were significant on all attributes and their inclusion significantly improved the fit of the model (χ2 P = 0.001). We estimated separate models for each site as well as the overall to aid in interpretation.