Postnatal infection surveillance by telephone in Dar es Salaam, Tanzania: An observational cohort study

Introduction Maternal and newborn infections are important causes of mortality but morbidity data from low- and middle-income countries is limited. We used telephone surveillance to estimate infection incidence and risk factors in women and newborns following hospital childbirth in Dar es Salaam. Methods We recruited postnatal women from two tertiary hospitals and conducted telephone interviews 7 and 28 days after delivery. Maternal infection (endometritis, caesarean or perineal wound, or urinary tract infection) and newborn infection (umbilical cord or possible severe bacterial infection) were identified using hospital case-notes at the time of birth and self-reported symptoms. Adjusted Cox regression models were used to assess the association between potential risk-factors and infection. Results We recruited 879 women and interviewed 791 (90%). From day 0-7, 6.7% (49/791) women and 6.2% (51/762) newborns developed infection. Using full follow-up data, the infection rate was higher in women with caesarean childbirth versus women with a vaginal delivery (aHR 1.93, 95%CI 1.11-3.36). Only 24% of women received pre-operative antibiotic prophylaxis before caesarean section. Infection was higher in newborns resuscitated at birth versus newborns who were not resuscitated (aHR 4.45, 95%CI 2.10-9.44). At interview, 66% (37/56) of women and 88% (72/82) of newborns with possible infection had sought health-facility care. Conclusions Telephone surveillance identified a substantial risk of postnatal infection, including cases likely to have been missed by hospital-based data-collection alone. Risk of maternal endometritis and newborn possible severe bacterial infection were consistent with other studies. Caesarean section was the most important risk-factor for maternal infection. Improved implementation of pre-operative antibiotic prophylaxis is urgently required to mitigate this risk. This study was a collaboration between London School of Hygiene and Tropical Medicine (LSHTM) and Ifakara Health Institute (IHI) and based at two of the three public Regional Referral Hospitals in Dar es Salaam; Amana (Ilala district) and Temeke (Temeke district). Each hospital conducts approximately 1,000 births per month. It was a sub-study of a pilot evaluation of training in environmental cleaning [15]. Two research nurses per hospital recruited eligible women from postnatal wards every Monday to Thursday. They sampled from all women who gave birth in the previous 24 hours using a random number application [16] with probability proportional to delivery mode (caesarean or vaginal). Eligible women were aged 18 years or older with access to at least one mobile telephone and providing signed or witnessed thumbprint consent. Women admitted to the intensive care unit were ineligible. Women provided up to three mobile telephone numbers; one or two of their own and one for a relative or neighbour. Replacements were sampled in the same way when potential participants were unavailable or ineligible. Two research nurses at IHI offices in Dar es Salaam interviewed each woman twice by telephone in Kiswahili, starting seven and 28 days after recruitment. Nurses made four telephone call attempts, over seven days, to reach each woman. The primary outcomes were 1) possible maternal postnatal infection (one or more of caesarean surgical site infection, urinary tract infection, perineal wound infection or endometritis) and 2) possible newborn infection (either of pSBI or umbilical cord infection). Each outcome was measured as a rate, and as the day 7 (early infection) and day 8–28 cumulative risk. Infections were identified from women’s hospital case-notes around the time of childbirth or from self-reported symptoms during telephone interview using standard definitions [7, 17, 18]. These definitions were adapted by the first author to include only symptoms and signs easily reported by the women (Table 1). Secondary outcomes were each individual infection listed above, plus mastitis. a)Centres for Disease Control [18] b) Scottish Intercollegiate Guidelines Network [17] c)Young Infants Clinical Signs Study [7]. Potential risk factors were extracted from hospital case-notes; maternal age, gestational age, parity, HIV, diabetes, hypertensive disorder, haemorrhage, prelabour rupture of membranes (PROM), induction of labour, delivery mode, postpartum haemorrhage (PPH) and infection during labour. Possible consequences of infection collected during telephone interview were self-reported readmission, depression assessed using a validated 5-question modified Edinburgh Postnatal Depression Scale (EPDS) and functionality according to five common postpartum activities (S1 Appendix). Data was entered on tablets with Open Data Kit (ODK), using unique identification (ID) numbers to maintain confidentiality. Data was extracted from maternal paper case-notes after hospital discharge, including demographics, pregnancy and childbirth history, infection diagnosed during admission and antibiotics prescribed (S2 Appendix). Telephone interviews with women consisted of pre-coded closed questions on the history of specific symptoms of infection, day of symptom onset, care-seeking behaviour, and readmission to hospital. At day-28, women were also asked questions on depression and function (S1 Appendix). Women with infection symptoms were advised to attend a health-facility if they hadn’t already. In cases of maternal depression or neonatal death, women were offered referral to social welfare liaison for counselling and support. Research nurses received six days training in recruitment and data collection, including two days at the hospitals when they piloted the tools on 24 women. Telephone interview nurses additionally conducted pilot interviews with the same 24 women over two days. With 900 women and an estimated 10% loss to follow-up at day-28, we would have 95% confidence to estimate a maternal infection risk of 3%±1.2% with 80% power. Our daily recruitment target was 12–20 women per hospital. Data was cleaned and analysed using STATA 16. Gestational age was grouped as preterm (<37 weeks) or term (37–42 weeks). The depression score was grouped as no depression (0–5) or possible depression (6–30). Maternal function questions were analysed individually as “any” or “no difficulty” in performing the function. Duplicate ID numbers and data entry errors were corrected where possible using hospital case-notes or comparing with other study data. Any remaining discordant data was dropped. There was inconsistency in the occurrence of stillbirths between data sources, therefore stillbirths were not analysed. Data on twin and triplet newborns was also inconsistent and in addition an error in ODK programming meant only data from the first baby was useable. Women’s demographic and pregnancy data was described by delivery mode. Rates of infection were calculated from delivery until the day-28 telephone call using reported days from delivery to start of symptoms. Symptoms reported at both day-7 and day-28 were counted as distinct infection events if they started over 14 days apart, or if they met criteria for different infection types and started over seven days apart, or if initial symptoms had resolved by the day-7 interview. Date of death and infection data were not collected from babies who died before the day-7 interview, therefore these babies were excluded from infection outcome analyses. Babies who died after the day-7 interview contributed to infection analyses up to day 7. Using Cox regression with robust standard errors to account for clustering by person, we explored associations between potential risk factors and the rate of maternal postnatal infection or possible newborn infection. Proportional hazards assumptions were checked using tests based on Schoenfeld Residuals. Factors showing evidence of association in the crude analysis (p<0.1) were explored further in multivariable models. Maternal age and delivery hospital were considered a priori confounders for risk of maternal postnatal infection. We restricted the parameters in the final models to 10% of the number of outcomes. For missing risk-factor data, we carried out multiple imputation using chained equations (MICE) because most variables were categorical, creating 10 imputed datasets. Delivery mode and hospital were included as auxiliary variables. Women whose case-notes were missing were excluded from risk-factor analysis. We report the highest level of care sought by women and newborns with possible infection and the percentage readmission to hospital for those with and without infection. We describe maternal depression and function at day-28 and explore associations with early postnatal infection using chi-squared tests and logistic regression. The study was approved by the Tanzanian National Institute for Medical Research, IHI Institutional Research Board and LSHTM Research Ethics Committee. Written informed consent was obtained from women on the postnatal wards. Willingness to continue in the study was confirmed at the start of each telephone interview. There was no public or patient involvement in the study design or interpretation of results. The Soapbox Collaborative supported the study following external peer review of the study proposal.