Menu
Menu
Menu
Menu
Objectives. We fielded a population-based discrete choice experiment (DCE) in rural western Tanzania, where only one third of women deliver children in a health facility, to evaluate health-system factors that influence women’s delivery decisions. Methods. Women were shown choice cards that described 2 hypothetical health centers by means of 6 attributes (distance, cost, type of provider, attitude of provider, drugs and equipment, free transport). The women were then asked to indicate which of the 2 facilities they would prefer to use for a future delivery. We used a hierarchical Bayes procedure to estimate individual and mean utility parameters. Results. A total of 1203 women completed the DCE. The model showed good predictive validity for actual facility choice. The most important facility attributes were a respectful provider attitude and availability of drugs and medical equipment. Policy simulations suggested that if these attributes were improved at existing facilities, the proportion of women preferring facility delivery would rise from 43% to 88%. Conclusions. In regions in which attended delivery rates are low despite availability of primary care facilities, policy experiments should test the effect of targeted quality improvements on facility use.
Kasulu District, with a population of 630 000, lies in Kigoma Region in western Tanzania. It is a primarily rural district with 1 main town, Kasulu (population 33 000).23 There are 3 hospitals, 10 health centers, and 57 dispensaries. Government dispensaries are small primary care facilities with basic diagnostic equipment and 1 or 2 beds for deliveries. Dispensaries are staffed predominantly by clinical officers (trained to manage basic health conditions), and health centers are staffed predominantly by clinical officers, nurses, and assistant medical officers (clinical officers with additional training).24 The poor roads and unavailability of transport combined with a scarcity of referral hospitals obliges the population to rely mainly on primary care facilities for maternal health services. The population belongs primarily to the Muha tribe and speaks both Kiswahili and the local language, Kiha. We selected a representative cluster sample of rural households from Kasulu District, omitting the town of Kasulu. Fifty villages were chosen in the first stage, with probability proportional to size, on the basis of the 2002 Tanzania census. Within each village, 1 subvillage, each with approximately 100 households, was randomly selected. The leader of the selected subvillage provided a list of households within the subvillage from which 35 households were selected through random systematic sampling. Households in which there was a woman 18 years or older who had had a delivery in the previous 5 years were eligible for inclusion in the study. Written consent was obtained from all respondents. The DCE was designed to estimate the relative value or utility of different features of health facilities to women from Kasulu District in considering where to deliver their next child. Before administering the DCE, we used a standard questionnaire to obtain information about (1) the women’s sociodemographic characteristics; (2) their household material assets, such as animals, mosquito nets, bicycles, and type of roof (used to construct a measure of socioeconomic status); and (3) their past and planned future places of delivery. In designing the DCE, we selected attributes (features) of the service, assigned levels to each attribute, identified the scenarios to present, and fielded the experiment to establish preferences. On the basis of a review of literature on determinants of access to health services in sub-Saharan Africa, interviews with providers, and pretesting with rural women, we selected 6 policy-amenable facility attributes: distance, cost, provider attitude (a measure of responsiveness), availability of drugs and equipment (a measure of technical quality), type of provider, and transport. (A detailed description of the selection process and fielding is provided in the appendix available as an online supplement to this article at http://www.ajph.org.) It took approximately 30 minutes to administer the full interview, including the DCE. The interviews were administered from June to mid-July 2007. The responses were recorded with pencil and paper, entered into a text file, cleaned, and imported into SAS version 9.1.13 (SAS Institute Inc, Cary, NC) and Sawtooth software version 4.4.6 (Sawtooth Software, Inc, Sequim, WA). A sample DCE card is shown in Figure 1. Sample discrete choice experiment card and script presented to women (N = 1203) from Kasulu District: Kigoma, Tanzania, 2007. We calculated descriptive univariate statistics for demographic and place-of-delivery variables. We used SAS-callable SUDAAN to account for the survey’s cluster design.25 We used Sawtooth’s Choice-Based Conjoint with Hierarchical Bayes statistical program to estimate coefficients for the individual utilities of each attribute level (details in appendix available at http://www.ajph.org).26 Using market simulator software in Sawtooth’s Choice-Based Conjoint with Hierarchical Bayes module, we used individual-level utilities to estimate the proportion of respondents who would prefer specific facility profiles.26 The simulations calculate total utilities for the simulated facility for each respondent by summing attribute utilities. The respondents were repeatedly sampled to stabilize these preferences, and we added a random error term to the estimates of utilities to correct for any similarities in scenarios.27 We used the simulations to explore the predictive validity of the utility parameter estimates, both within the experiment and with real-life behavior. First, we compared predicted to actual facility choices within our experiment. To do this, we split our sample into 2 groups: women who received DCE versions 1 through 4 and women who received version 5, which we designated as the holdout scenarios. We estimated unstandardized parameter estimates for facility attribute levels only for respondents who were given DCE versions 1 through 4. Using these estimates, we then calculated the aggregate utility of alternative A and alternative B in each of the version 5 holdout scenarios. The facility with the higher aggregate utility was identified as the preferred facility. These predicted preferences were then compared with actual selections of preferred facilities by women who were given version 5. Second, to assess the predictive validity of our model for real-life behavior (revealed preference), we modeled the predicted percentage of women who would choose to deliver in a currently available health facility versus at home. We did this by assigning attribute levels corresponding to the current state of dispensaries, health centers, and hospitals available to women in Kasulu District and to attributes of home delivery, on the basis of information from local health providers.28 We compared the predicted proportion of home and facility deliveries in the model to the actual place of delivery for women’s most recent child (on the basis of this survey) and to results from the previous 2 national Demographic and Health Surveys.11,29 To assess the extent to which focused investments and policy reforms in the Tanzanian health system would increase utilization of the most widely available facilities—in this case dispensaries—we conducted simulations reflecting potential changes to facilities, and we calculated projected shares of women’s preference for these facilities versus delivering at home. (Assumptions used in the policy simulations are shown in the appendix available at http://www.ajph.org).
N/A
