Logistic regression as a prognostic tool for estimating development of postpartum endometritis and lactostasis

Abstract

Introduction: During our investigation of postpartum purulent-inflammatory diseases we implied both standardised set of diagnostic tools as well as fluorescent spectroscopy. ROC-analysis and logistic regression were implied to perform a prognostic assessment of the development of postpartum endometritis and lactostasis. Objectives: We aim to assess prognosis of development of postpartum endometritis and lactostasis among postpartum women. Methods: General clinical assessment methods, biochemical and instrumental analysis, including fluorescent spectroscopy and statistical assessment (logistic regression). The main cohort of participants consisted of 30 postpartum women with postpartum endometritis and lactostasis. The control group was formed from 40 postpartum women with uncomplicated post labour period. Results: We performed and implemented a stepwise logistic regression (with forward selection) aimed to separate the factors, whose cumulative effect would have a significant effect on development of postpartum endometritis and lactostasis. Through statistical analysis, we were able to determine 6 clinically significant factors (indication of colpitis, TORCH-infections, labour duration >12 hrs, number of bed days and the maximum value of the wavelength in fluorescence intensity of blood serum derived from fluorescent spectroscopy) which cumulatively have an impact on occurrence of postpartum endometritis and lactostasis. Optimal threshold of decision making (cut-off point) which allows a maximum balance between sensitivity and specificity for this model was 0.47. Therefore, when either of these variables has a value of 0.47≤, we can conclude that the patient has a high risk of developing postpartum endometritis and lactostasis. Consequently, if the 0.47≥, we can confirm the absence of risk of postpartum endometritis and lactostasis development. Among 93.33% patients of the main cohort this indicator exceeded the optimal threshold value of 0.47. On the other hand, only 7.5% of patients from control group had an indicator of more than 0.47. Conclusion: As a result of our study, we proposed a prognostic model which allowed prognosis of development of postpartum endometritis and lactostasis. The resulting model is correct with a probability of >99% (р<0.001; χ² = 64.88; df = 6).