December 2006

1. SCHOOL OF HEALTH SYSTEMS AND PUBLICHEALTH UNIVERSITY OF PRETORIA Faculty of Health Sciences Tel: (012) 841 3240 Fax: (012) 841 3328  PO Box 667, Pretoria, 0001 http://shsph.up.ac.za Factors that affect utilization of modern contraceptives and family planning methods among women with ages 15 to 49 in and around the City of Tshwane (Pretoria), South Africa December 2006

2. Authors • Dr. Zeleke Worku School of Health Systems and Public Health University of Pretoria • Dr. Delphin Tshibangu City of Tshwane Metropolitan Council (CTMM), Pretoria

3. Introduction • Population size of Tshwane is 1.8 million • Women account for 50% of population • 57% of women are aged 15 to 49 years • Above 50% of black females are characterized by low income, high unemployment rate, unwanted pregnancies, miscarriages, abortions and poor access to family planning services

4. Figure 1: City of Tshwane Metropolitan Council (CTMM)

5. Introduction (ctd.) • 4 major health districts in CTMM • 20 electoral wards in CTMM • 57% of women are aged 15 to 49 years • Black females account for 64.7% to 68.7% • White females account for 28.1% to 30.1% • Indian and Coloured females account for 3.2% to 5.2% of the female population

6. Introduction (ctd.) • Disease pattern follows socio-economic status • Health services fragmented and skewed • The poor and jobless experience TB, HIV/AIDS, malnutrition, upper and lower respiratory tract infections • The rich experience cardio-vascular diseases such as high cholesterol and myocardial infection

7. FP facilities (11 types)

8. Background of problem • High teenage pregnancy, adverse pregnancy outcomes, underutilization of modern family planning methods and services, high prevalence of HIV and TB, high unemployment rate among poor women • Lack of reliable large sample studies assessing contraceptive use

9. Objectives of study • To assess the level of utilization of modern contraceptives by women 15-49 years of age • To identify key predictors of contraceptive use • To find out if there is variability among wards • To find out if there is variability among facilities

10. Study design and sample size • Cross sectional and descriptive study design conducted in 2004 by the CTMM • Three-stage cluster sampling with PPS (Probability Proportional to Size) was used • Data gathered from a total of 8, 497 women aged 15-49 living in 4 districts of the CTMM (Pretoria) • Analysis of 3, 547 women aged 15-49 and using modern FP methods (41.74%)

11. Outcome variable of study (Y)

12. Independent variables • Socio-economic • Demographic and fertility related • Health related • Variables related to sexual behaviour and practice of respondent • Variables related to sexual behaviour and practice of sexual partner

13. Methods of data analysis • Frequency tables for discrete variables • Summary stats for continuous variables • Factor analysis for data reduction • Binary logistic regression • Multilevel analysis (Proc mixed) based on a hierarchical nested design

14. Modern family planning methods • Injections • Pills • Condoms • Intra-Uterine Devices (IUDs) • Sterilization • Vaginal foam

15. Description of FP users • Number of FP users with ages 15 to 49 years = 3, 547 (41.74% of 8, 497) • Average age of FP users at first sex = 18.72 years • Average age of FP users at first pregnancy = 19.36 years • Prevalence of teenage pregnancy = 9.5% • Prevalence of adverse outcomes = 12.19%

17. Ages of FP users with jobs

18. Modern contraceptive use by district

19. Marital status of FP users

20. Level of education of FP users

21. Number of children of FP users

22. Distribution of modern FP methods

26. Figure 6: Desire for latest pregnancy

29. Figure 9: Distribution of condom use by level of education

31. Binary logistic regression analysis • n = 8, 497 women aged 15 to 49 • 3, 547 modern FP users (41.74%) • 4, 950 non-users (58.26%) • Traditional FP methods (withdrawal, abstinence, herbs, etc) are not included in modern FP methods Objective • To identify key predictors of use of modern FP methods

32. Variables used for logistic regression Y: Use of modern FP method (1, 0) List of predictor variables • Age at first sex • Age at first pregnancy • Employment • Awareness about FP methods and services • Access to FP services • Level of education • Level of income

33. List of variables (ctd.) • Employment • Nearby clinic • Religion • Family size • Number of children • Support from partner • Sexual behaviour and practice of woman • Sexual behaviour and practice of partner

34. Adjusted Odds Ratios from logistic regression (1) ________________________________________ Variable OR 95% CI P-value ________________________________________ Family size Less than 5 4.89 (2.14, 6.03) 0.0000 5 or above 1.00 Nearby community health service center Available 6.44 (3.89, 8.19) 0.0000 Not available 1.00

35. Adjusted Odds Ratios from logistic regression (2) ________________________________________ Variable OR 95% CI P-value ________________________________________ Age at first sex Less than 19 2.06 (1.13, 3.42) 0.0411 19 or above 1.00 Age at first pregnancy Less than 20 1.44 (0.74, 2.26) 0.0613 20 or above 1.00

36. Confounding variables ________________________________________ Variable OR 95% CI P-value ________________________________________ Level of education Primary or less 1.27 (0.78, 1.68) 0.1541 Secondary or above 1.00 Employment ever Not employed 1.21 (0.79, 1.64) 0.1557 Employed 1.00

37. Effect modifying variables ________________________________________ Effect modifiers (Significant interaction effects) District by education 4.77 (1.89, 6.33) 0.0003 District by job 3.16 (1.09, 5.01) 0.0004 Family size by job 3.51 (1.41, 5.49) 0.0004

38. Multilevel analysis (1) n = 3,547 modern FP users only Research questions • Does FP choice vary by facility? • Does FP choice vary by ward? • Are age at first sex, availability of nearby FP services, family size, level of education and employment status influential over FP choice?

39. Multilevel analysis (2) Y: Type of modern FP method (1, …., 6) Level 1 is the individual woman (1, .., 3547) Level 2 is FP facility (1, …, 11) Level 3 is ward (1, …, 20) List of predictor variables Age at first sex Availability of nearby FP services Family size Level of education Employment status

40. Multilevel analysis (3) • Hierarchical nested design model • Women nested within facilities • Facilities nested within wards • The variable facility was assumed to be random because the number of facilities per ward was 11 > 10 • Variability among wards and FP facilities compared using Intra Class Correlation (ICC) • Adjustment made for clusters

41. Multilevel analysis (4) Main Advantage: • FP services are hierarchical • FP users are nested within facilities, and facilities are nested within wards • Traditional multiple linear regression models (OLS) underestimate the standard error of estimation because the assumption of independence of observations is violated

42. Model for 3-level analysis

43. Maximum Likelihood Estimator Let ZU + = combined error term  Then, is distributed multivariate normal with mean and an variance–covariance matrix V = Let ZU + = combined error term Then, is distributed multivariate normal with mean and an variance –covariance matrix V = Let ZU + = combined error term Then, is distributed multivariate normal with mean and an variance –covariance matrix V =

44. Restricted MLE estimator Let ZU + = combined error term Then, is distributed multivariate normal with mean and an variance –covariance matrix V = Let ZU + = combined error term Then, is distributed multivariate normal with mean and an variance –covariance matrix V =

45. Intra Class Correlation (ICC) Let ZU + = combined error term  Then, is distributed multivariate normal with mean and an variance–covariance matrix V = Let ZU + = combined error term Then, is distributed multivariate normal with mean and an variance –covariance matrix V = Let ZU + = combined error term Then, is distributed multivariate normal with mean and an variance –covariance matrix V =

46. Intra Class Correlation (ctd.) Let ZU + = combined error term  Then, is distributed multivariate normal with mean and an variance–covariance matrix V = Let ZU + = combined error term Then, is distributed multivariate normal with mean and an variance –covariance matrix V = Let ZU + = combined error term Then, is distributed multivariate normal with mean and an variance –covariance matrix V =

47. Properties of ICC Let ZU + = combined error term Then, is distributed multivariate normal with mean and an variance –covariance matrix V = Let ZU + = combined error term Then, is distributed multivariate normal with mean and an variance –covariance matrix V =

48. Properties of ICC (ctd.) Let ZU + = combined error term Then, is distributed multivariate normal with mean and an variance –covariance matrix V = Let ZU + = combined error term Then, is distributed multivariate normal with mean and an variance –covariance matrix V =

49. Scatter plot of FP method by AFS Let ZU + = combined error term Then, is distributed multivariate normal with mean and an variance –covariance matrix V = Let ZU + = combined error term Then, is distributed multivariate normal with mean and an variance –covariance matrix V =

50. Covariance Parameter Estimates ___________________________________ Variance Estimate P-value ___________________________________ Ward 0.06471 0.0104 Facility 0.08802 0.0101 Error 1.89200 0.1763 Total variance 2.0447