2020年2月5日    担当：ソン

＜論文の要約＞

【Introduction】

Anemia is a growing public health concern worldwide. Exposure to anemia has been shown to be related to adverse maternal and perinatal outcomes. To date there has been little agreement on the robustness of published studies’ results because of their methodical limitations. By using standardized procedures, this study attempted to investigate the association between severe anemia and maternal mortality.

【Methods】

Survey methods and participants Data were drawn from the WHO Multicountry Survey which was a cross-sectional study and was based on a multi-stage sampling method. The first stage of sampling was the selection of countries. The second stage consisted of randomly selected two provinces in addition to the capital city within each country. Next eligible institutions were sampled. Participants were recruited from 359 health facilities in 29 countries across Latin America, Africa, the Western Pacific, eastern Mediterranean and southeast Asia from May 2010 to December 2011. All women giving birth and their expected newborns, all maternal near miss cases, and all maternal deaths were included.

Definitions and variables The outcome was maternal death which was defined as death any time after admission up to seven postpartum days. Severe anemia was defined as Hb < 70 g/L. The covariates including clinical characteristics, clinical indicators, and laboratory markers were used to establish the propensity score.

Statistical analysis In order to examine the association between severe anemia and maternal death, a multilevel logistic regression and a propensity score regression were performed. The multilevel logistic regression had three levels where participants were nested within facilities and facilities within countries. Selected covariates had a reasonable biological relation to the association between severe anemia and maternal death. Confounding effect was defined as a change of greater than 10% between the adjusted and unadjusted odds ratio for severe anemia and maternal death. A 1:2 propensity score matching model with a caliper of 0.25 of the standard deviation of the propensity score was also used in order to draw the causal inference. Standardized differences and C statistic were used to assess the degree of balance. The effect of reductions of severe anemia upon maternal death was then analyzed using the population attributable fraction. All analyses were carried out using R and STATA software.

【Results】

Of the study population, 312281 participants (with 4687 cases of severe anemia) were included in the multilevel analysis, and 12470 women (4189 women with severe anemia were matched with 8981 women without severe anemia) along with 146 maternal deaths were included in the propensity score analysis. There was a significant positive association between severe anemia and maternal death in both multilevel analysis and propensity score analysis. The results of sensitivity analyses (when cases of post-partum hemorrhage in the multilevel regression model were removed or reclassified and when the caliper of the propensity score was changed) were similar to those of the primary analysis.

【Discussion】

Overall, these results indicated that women with severe anemia had 2.36 times higher odds of death than women without severe anemia. This was contrary to the assumption that hemoglobin is not a suitable measure of outcome. The strengths of the study included its large and diverse dataset; its statistical methods with a combination of two types of regression analyses; and its chosen covariates which had a plausible biological relation with the outcome and main dependence variable. The study had several limitations. First, actual concentrations of hemoglobin and timing of severe anemia were not recorded. Second, over 4000 cases that contained missing values were excluded without a thorough examination. Other sources for possible error were that potential confounding could not be completely excluded and a stratified analysis by individual facilities was not done. Being strong, consistent, and robust in both multilevel and propensity score regression analysis, however, the findings suggest that healthcare professionals should give precedence to maternal anemia with effective prevention strategies.

＜ジャーナルクラブでのディスカッション＞

■Why do we need multilevel analysis and propensity score matching techniques?

■What is caliper?

■What is C statistic?

■What is population attributable fraction?

■Finally, this paper comes into my head when I think up ideas for JECS study. So, I wish we will hit upon some great ideas.

Please attending and sharing with me your knowledge and experiences. And do not forget to have many questions for our teachers so that we can learn a lot from them.