Quality of care and maternal mortality in a tertiary-level hospital in Mozambique: a retrospective study of clinicopathological discrepancies

Background: Although an increasing number of pregnant women in resource-limited areas deliver in health-care facilities, maternal mortality remains high in these settings. Inadequate diagnosis and management of common life-threatening conditions is an important determinant of maternal mortality. We analysed the clinicopathological discrepancies in a series of maternal deaths from Mozambique and assessed changes over 10 years in the diagnostic process. We aimed to provide data on clinical diagnostic accuracy to be used for improving quality of care and reducing maternal mortality. Methods: We did a retrospective analysis of clinicopathological discrepancies in 91 maternal deaths occurring from Nov 1, 2013, to March 31, 2015 (17 month-long period), at a tertiary-level hospital in Mozambique, using complete diagnostic autopsies as the gold standard to ascertain cause of death. We estimated the performance of the clinical diagnosis and classified clinicopathological discrepancies as major and minor errors. We compared the findings of this analysis with those of a similar study done in the same setting 10 years earlier. Findings: We identified a clinicopathological discrepancy in 35 (38%) of 91 women. All diagnostic errors observed were classified as major discrepancies. The sensitivity of the clinical diagnosis for puerperal infections was 17% and the positive predictive value was 50%. The sensitivity for non-obstetric infections was 48%. The sensitivity for eclampsia was 100% but the positive predictive value was 33%. Over the 10-year period, the performance of clinical diagnosis did not improve, and worsened for some diagnoses, such as puerperal infection. Interpretation: Decreasing maternal mortality requires improvement of the pre-mortem diagnostic process and avoidance of clinical errors by refining clinical skills and increasing the availability and quality of diagnostic tests. Comparison of post-mortem information with clinical diagnosis will help monitor the reduction of clinical errors and thus improve the quality of care. Funding: Bill & Melinda Gates Foundation and Instituto de Salud Carlos III. This retrospective study was done at the Maputo Central Hospital (Maputo, Mozambique), a 1500-bed government-funded tertiary-level health-care facility. Recruitment of maternal deaths was done from Nov 1, 2013, to March 31, 2015 (17-month period). All deceased women who fulfilled the standard WHO definition of a pregnancy-related death,12 and for whom the family had given verbal informed consent for the autopsy requested by the clinician, were included. Accidental or incidental deaths were excluded. Following the guidelines of the Ministry of Health of Mozambique, all maternal deaths occurring at the Maputo Central Hospital undergo a complete diagnostic autopsy unless the family does not provide consent. This study received approval from the National Bioethics Committee of Mozambique (342/CNBS/13) and the Clinical Research Ethics Committee of the Hospital Clinic of Barcelona (Spain; 2013/8677). A complete dissection was done with macroscopic evaluation of all organs according to a standardised protocol.13 Samples of grossly identified lesions and of solid organs, including the uterus, were collected for histological examination; additionally, samples of blood and cerebrospinal fluid were obtained. When available, the placenta was macroscopically evaluated and sampled. Histological evaluation comprised staining with haematoxylin and eosin in all samples and additional histochemical or immunohistochemical stains (eg, Ziehl-Neelsen or Plasmodium falciparum immunohistochemical staining) when needed. The extensive microbiological analysis done has been reported in detail elsewhere.14 Briefly, universal screening was done, which comprised detection of P falciparum by PCR, detection of antibodies against HIV-1 and HIV-2 and HIV viral load, and bacterial or fungal cultures of blood and cerebrospinal fluid. Additional microbiological screening was applied to HIV-positive cases, including real time PCR in cerebrospinal fluid for Toxoplasma gondii, Mycobacterium tuberculosis, and Cryptococcus spp and real-time PCR in lung samples for Pneumocystis jirovecii. Molecular methods were used in cases in which the histological features were discordant with the culture results (eg, pneumonia by histology and no infectious agent identified on culture). Patient data, including demographic information, previous medical history, and inpatient admission process (collected by clinicians in charge, including obstetricians) were extracted from medical records and recorded in a standardised questionnaire by a study medical doctor (QB). Up to five clinical diagnoses registered in medical records by the caring clinicians were selected and abstracted. The first diagnosis listed was regarded as the main diagnosis, and the remaining diagnoses were classified as secondary. Macroscopic, microscopic, and microbiological findings of complete diagnostic autopsies and any available clinical information were evaluated by a panel of multidisciplinary experts that comprised clinical (maternal and child health) and laboratory (pathology and microbiology) specialists, and the final complete diagnostic autopsy diagnosis was assigned. As previously described,14 all morbid conditions directly leading to death, any underlying conditions, and any other clinically significant conditions possibly contributing to death were classified as either direct obstetric or indirect obstetric deaths, and codified according to the International Classification of Diseases, 10th revision.12, 15 Diseases were grouped into the following eight categories: (1) pregnancies with abortive outcome; (2) hypertensive disorders in pregnancy, childbirth, and puerperium; (3) obstetric haemorrhage; (4) pregnancy-related infections; (5) other obstetric complications; (6) unanticipated complications of management; (7) non-obstetric complications; and (8) unexplained deaths. We considered categories 1 to 6 direct obstetric deaths, whereas category 7 was considered to correspond to indirect obstetric deaths. When more than one severe diagnosis was identified, the disease most likely to have caused the death was considered the final complete diagnostic autopsy diagnosis.14 Diagnostic discrepancies were classified as major or minor.16, 17 Major discrepancies involved major diagnoses and were classified as class I or class II. Class I refers to discrepancies in which the knowledge of the correct diagnosis before death would have led to changes in clinical management that could have prolonged survival or cured the patient (eg, pyogenic meningitis treated as eclampsia). In class II errors, patient survival would have not been modified (eg, fulminant hepatitis treated as sepsis). Minor discrepancies involved minor diagnoses and were classified as class III (non-diagnosed diseases with symptoms that should have been treated—eg, mild aspiration pneumonia in a patient with eclampsia) and class IV (non-diagnosed diseases with possible epidemiological or genetic importance—eg, schistosomal infections). Correctly diagnosed patients were classified as class V. Class VI comprised non-classifiable cases (autopsy unsatisfactory or with no clear diagnosis). For analysis of clinicopathological discrepancies, two masked investigators assessed each case; their evaluations were compared and a third rater evaluated any discrepant cases. The following information was provided to each rater: autopsy final diagnosis, antecedent causes, and other significant conditions and clinical diagnoses (main diagnosis, and up to a maximum of four additional diagnoses) extracted from the medical record. Clinicopathological correlation was determined by assessing whether the complete diagnostic autopsy diagnosis was identified among any of the clinical diagnoses. A case was considered discrepant when there was no coincidence between any of the five clinical diagnoses listed by the clinician and the final cause of death identified in the complete diagnostic autopsy. In each case, only the worst diagnostic error was considered. We did a comparative analysis of the performance of the clinical diagnosis of four main maternal death categories between the current findings and those of a study undertaken 10 years earlier in the same hospital and using the same methods to determine cause of death.11 We assessed concordance between raters with the κ statistic.18 We compared proportions by χ2 test and used logistic regression with penalised likelihood to evaluate factors associated with major clinical errors.19, 20 We used penalised likelihood to mitigate the bias caused by rare events in the dataset, as major errors were infrequent or non-existent for some covariates included in the analysis of associations or a combination of them in multivariable analyses (eg, ectopic gravidity, bloody diarrhoea, and choluria). This situation is referred to as separation or monotone likelihood and produces infinite estimates for some coefficients. In such a situation, it can be useful to maximise Firth’s penalised likelihood, rather than the usual likelihood.19 We calculated the sensitivity, specificity, positive predictive value, and negative predictive value for each diagnosis. We defined false-negative diagnoses as discrepancies for which the autopsy diagnosis was in the assessed diagnostic category, but the clinical diagnosis was in another diagnostic category. We defined false-positive diagnoses as discrepancies for which the clinical diagnosis was in the diagnostic category but not the autopsy diagnosis. We estimated a multivariable adjusted model using all covariates with p≤0·15 in the crude analysis. Data were analysed with STATA (version 15). The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.