Impact of COVID-19 pandemic on time series of maternal mortality ratio in Bahia, Brazil: analysis of period 2011–2020

Background Most studies on the effects of SARS-CoV-2 infection have been conducted with adults and non-pregnant women. Thus, its impacts on maternal health are not yet fully established. This study aimed to verify the relationship between the maternal mortality ratio and the incidence of COVID-19 in the State of Bahia, Brazil, 2020. Methods This time-series study used publicly available information in Brazil, to obtain data on maternal deaths and live births in Bahia, State, from January 1, 2011, to December 31, 2020. The time trend of Maternal Mortality Ratio (MMR) was analysed through polynomial regression, of order 6. Expected MMR, monthly (Jan-Dec) and annual values for 2020, were predicted by the additive Holt-Winters exponential smoothing algorithm, with 95% confidence interval, based on the time series of the MMR from 2011 to 2019, and the accuracy of the forecasts for 2020 was assessed by checking the smoothing coefficients and the mean errors. According to the statistical forecast, the MMR values recorded in the year 2020 were compared to those expected. Results In 2020, the annual MMR in Bahia, Brazil, was 78.23/100,000 live births, 59.46% higher than the expected ratio (49.06 [95% CI 38.70–59.90]). The increase in maternal mortality ratio relative to expected values was observed throughout the 2020 months; however, only after May, when the COVID-19 epidemic rose sharply, it exceeded the upper limit of the 95% CI of the monthly prediction. Of the 144 registered maternal deaths in 2020, 19 (13.19%) had COVID-19 mentioned as the cause of death. Conclusions Our study revealed the increase in maternal mortality, and its temporal relationship with the incidence of COVID-19, in Bahia, Brazil, in 2020. The COVID-19 pandemic may be directly and indirectly related to this increase, which needs to be investigated. An urgent public health action is needed to prevent and reduce maternal deaths during this pandemic, in Brazil.

The clinical and epidemiological profile observed, at the beginning of the epidemic in China, showed that elderly (60 years old or more) with comorbidities such as diabetes, hypertension, heart disease, obesity, pneumopathies and immunosuppressed individuals were at higher risk for complications and deaths for COVID19, as has been observed in many countries affected by this pandemic [3][4][5][6]. At earlier stages of the epidemic, pregnant women without comorbidities were not considered at greater risk for COVID-19 and its related complications. However, with the greater spread of this disease, some authors began to describe the occurrence of severe forms, as well as abortion and deaths, even among pregnant women without comorbidities [7,8].
These findings have been reported in many countries such as the United States, United Kingdom, France, Mexico and Brazil [2,[9][10][11][12], and contributed to subsidise PAHO, to issue an Alert in August 2020, encouraging member countries to redouble efforts to ensure access and continuity of prenatal care, with special attention to the early detection of signs, symptoms, and severity of clinical manifestations of COVID-19 [13]. It is known that during pregnancy the women present a relative immunodeficiency and it could worsen the clinical evolution of COVID 19 and lead to negative outcomes in the mothers and foetus [14].
Brazil is a middle-income country where the maternal mortality is still high [15] although it has been presenting a slow declining trend [16,17]. Since the 1990 s, improvements in the living and health conditions of the population in this country, which includes women of reproductive age, have contributed to an important reduction in maternal deaths [18], but not enough to reach the Brazilian government target of less than 30 maternal deaths for every 100,000 live births [19]. Then, the COVID-19 pandemic may negatively influence the evolution of this indicator in Brazil.
Given the uncertainties about the possible impact of the COVID-19 on maternal mortality, this study aimed to verify the relationship between the maternal mortality ratio and the incidence of COVID-19 in the State of Bahia, Brazil, 2020.

Methods
A retrospective time series study was carried out on maternal mortality during the COVID 19 pandemic in Bahia, Brazil, 2020. Maternal death was defined as established by the World Health Organization's International Classification of Diseases (ICD-10) [20].
The website of the Bahia State Health Secretariat [21] was accessed to obtain the data for this study. Epidemiological bulletins [22] were the sources of data on the number of new cases of COVID-19 and population estimate, in Bahia. According to the month of occurrence, the number of maternal deaths and live births, from January 2011 to December 2020, were extracted from the public domain health open data, DATASUS [23,24]. The DATASUS aggregates health data from different Brazilian official information systems. Data on maternal deaths come from the Mortality Information System (SIM), which in 2011 reached 96.1% nationwide coverage [25]. Data on live births come from the Live Birth Information System (SINASC), which in 2010 has coverage of 94,8% in Brazil, and most recent studies in large Brazilian cities point to 100% coverage [26,27]. All maternal deaths occurred in 2020 were included in our analyses.
We estimated maternal mortality ratio (MMR) by dividing the number of maternal deaths by the number of live births, multiplied by 100,000, for each month and year [28]. Annual (from 2011 to 2020) and monthly (from January to December 2020) MMR were plotted on a time curve. The COVID-19 incidence in Bahia, per month in 2020, was calculated by the ratio between the number of new cases of this disease and the population estimate, followed by the multiplication by 100,000.
The time trend of MMR in the period from 2011 to 2020 was verified through polynomial regression analysis, of order 6. The additive Holt-Winters exponential smoothing model (ETS A, A, A), which considers the additive parameters error, trend and seasonality, was used to analyse the MMR time series 2011-2019 and predict the values for 2020, with a 95% confidence interval. Then, the real MMR values recorded in Bahia in 2020 were compared to those expected. The accuracy of the monthly and yearly MMR forecasts for 2020 was assessed by checking the Alpha, Beta and Gamma smoothing coefficients and the values of Scaled Mean Absolute Error (MASE), Symmetric Mean Percentage Absolute Error (sMAPE), Mean Absolute Error (MAE) and Root of the Mean Quadratic Simulation Error (RMSE).
As this study was carried out using public domain data, it was not necessary to submit it to the Research Ethics Committee.

Discussion
This study shows the rise in maternal mortality ratio in Bahia in 2020, insofar the officially recorded maternal deaths are far higher than the expected number for this year. A temporal relationship with COVID-19 pandemic was observed, since months in which the maternal mortality ratio exceeded the predicted value coincided with those with the highest incidence of COVID-19 in the state. The small number of maternal death certificates with COVID-19 diagnosed among the causes of death, which alone do not justify the excess of deaths observed in 2020, and leads us to consider the possibility of under diagnoses, since Brazil has not implemented universal testing of pregnant women [8] for COVID-19, and potential indirect effects of the pandemic. Maternal deaths attested due to not specified or ill-defined causes, related to ICD-10 O98.5-Other viral diseases complicating pregnancy, childbirth and the puerperium, for example, J18.9-Pneumonia, of unspecified aetiology or U04.9-Severe acute respiratory syndrome [SARS] unspecified, can mask undiagnosed cases of COVID-19.
Maternal deaths could result from maternal illness directly related to COVID-19 infection or an indirect effect of health service disruptions, and other indirect effects caused by the pandemic. First, maternal deaths directly related to COVID-19 in Brazil has been alarmed high [8]. However, it is also important to consider the possible indirect contributions of the pandemic in the occurrence of these deaths, as many pregnant women have stopped to attend antenatal appointments most likely because they did not feel protected against SARS-CoV-2 infection. Also, there may have been delay or resistance from some maternities with a lower level of complexity, to provide care to pregnant women with flu-like symptoms, or difficulty in the transportation of pregnant women to health units of a higher level of complexity located in other municipalities. In addition, some of these maternal deaths can still be under epidemiological investigation and analysis by the Epidemiological Surveillance Services and the State Maternal Mortality Studies Committee, so, the final cause of death can still be changed, and some deaths which COVID -19 has not been mentioned could emerge after evaluation. Lastly, the change in the organization of prenatal services during the pandemic, with a limited number of medical care provided to avoid crowding in the waiting rooms, may have caused delays in care, contributing to maternal deaths that could otherwise be avoided. Maternal mortality is an indicator of access and quality of women's health care [19], and this increase in maternal mortality showed in 2020 represents an unacceptable setback, that needs to be better clarified and faced. We understand that the public and private health sectors face many challenges in ensuring access and adequate care for all pregnant women. However, although the impact of the COVID-19 pandemic may affect many of them, it must be more remarkable for the poorest women, as occur already in the maternal mortality from other causes, to whom women of greater socioeconomic vulnerability are also more affected [18,29,30]. Therefore, the COVID-19 pandemic may exacerbate the social inequalities and injustices that already exist in Brazil [31,32].
This study has limitations inherent to the use of i) secondary data; ii) relatively small number of maternal deaths, that makes the time series more susceptible to variations, and; iii) preliminary data on maternal deaths in 2020, due to possible delays in feeding official information systems. In addition, as time series studies correspond to a subtype of aggregate (or ecological) studies, they are vulnerable to ecological fallacy. Consequently, its results cannot be inferred for the individual level. Added to this the fact that it is a descriptive study, and thus a cause-effect relationship cannot be inferred from its results. Another limitation refers to the forecasting. In general, in the exponential smoothing of time series, the parameter ranges and initial values are arbitrary, and this can impair the accuracy of the forecast. In our study, it was possible to observe that the confidence interval for monthly forecasts was wider, indicating that there was less precision in these, when compared to the annual estimates. The model applied in