The following is a summary of “Noninvasive prediction models of intra-amniotic infection in women with preterm labor,” published in the JANUARY 2023 issue of Obstetrics and Gynecology by Cobo, et al.

Women with intra-amniotic infections had the highest risk of premature birth and the worst outcomes among preterm labor patients. Amniocentesis, seen as overly intrusive by women and doctors, was necessary to diagnose intra-amniotic infection. To concentrate early efforts on high-risk preterm labor women while avoiding needless treatments in low-risk preterm labor women, noninvasive approaches for diagnosing intra-amniotic infection and/or early delivery were essential. The highest-performing models were used in the investigation to predict a composite result of intra-amniotic infection and/or spontaneous delivery within 7 days by fusing biochemical data with clinical and ultrasound information.

Data from a cohort of women admitted to the Hospital Clinic and Hospital Sant Joan de Déu, Barcelona, Spain, between 2015 and 2020 with a diagnosis of preterm labor at <34 weeks of gestation and who underwent amniocentesis to confirm or exclude intra-amniotic infection or inflammation were used. A transvaginal ultrasound was done at admission, and samples of the mother’s blood and vagina were taken. Vaginal proteins were investigated to predict the composite result using multiplex immunoassay, amino acids were investigated using high-performance liquid chromatography, and bacteria were investigated using 16S ribosomal RNA gene amplicon sequencing. To construct prediction models using machine learning that could be used in a validation cohort, researchers chose ultrasound, maternal blood, and vaginal predictors that could be validated with quick diagnostic procedures.

The study comprised a cohort of 288 women who experienced preterm labor at or before 34 weeks of gestation, 103 (35%) of whom had a composite result of intra-amniotic infection and/or spontaneous delivery within 7 days. Cohorts for derivation (n=116) and validation (n=172) were separated from the sample. Notably, four prediction models were put forth, including the ultrasound transvaginal cervical length, maternal C-reactive protein, vaginal interleukin 6 (measured with an automated immunoanalyzer), vaginal pH (measured with a pH meter), vaginal lactic acid (measured with a reflectometer), and vaginal Lactobacillus genus (measured with quantitative polymerase chain reaction), with areas under the receiving operating characteristic curve ranging from 82.2% (95% CI, ±3.1%) to 85.2% (95% CI, ±3.1%), sensitivities ranging from 76.1% to 85.9%, and specificities ranging from 75.2% to 85.1%.

The study’s findings demonstrated how noninvasive techniques appropriate for point-of-care systems might identify high-risk preterm labor and significantly help clinical management and outcomes while enhancing resource usage and patient satisfaction.

Reference: ajog.org/article/S0002-9378(22)00581-6/fulltext