Results
One hundred and forty-four women with spontaneous labor were included (Figure S1), of whom ninety-six delivered at term and forty-eight preterm. Clinical variables are described in Table 1 andTable 2 . Subcategories of preterm birth, based on gestational age, were similar with respect to age, multiparity and history of PTB, with a mean gestational week of 35.20±1.10, 31.0±0.24, and 22.10±0.52 for late preterm, very preterm, and extremely preterm delivery, respectively. Likewise, we did not identify differences in stages of labor according to term and preterm delivery groups (Table S1 and S2).
Cytokine concentrations were similar between women who delivered at term and preterm (Figure 1A; Table S3). However, when we stratified by subcategories of preterm birth, we identified that women who delivered extremely preterm had significantly higher concentration of pro-inflammatory cytokine IFN-γ compared with those with late preterm, but interestingly, they were not different with those who delivered at term (Figure 1B; Table S3).
Since we did not identify differences of cytokine concentrations in CVF between women who developed term and preterm labor, we analyzed them as a single group for every stage. We compared cytokine concentrations between stages of labor (Figure 1C; Table S4), and identified that pro-inflammatory cytokines, excluding IL-2, and the anti-inflammatory cytokine IL-10, increased in correlation with the sequence of spontaneous labor. The main concentration changes occurred with respect to Stage 0 and the final stages (Stage 3 and 4). In particular, IL-6 concentration showed a clear peak at Stage 4. We did not identify a clear trend in the concentrations of chemotactic cytokine IL-8, the pro-inflammatory cytokine IL-2, and the anti-inflammatory cytokines IL-4 and IL-1RA.
To test the hypothesis that CVF cytokines may characterize the initial clinical manifestations of spontaneous labor regardless of gestational age, we performed PCA of cytokines concentrations that showed a rising trend across labor stages. We identified a pro-inflammatory profile defined by IFN-γ, TNF-α, IL-1β and IL-6, and we extracted two principal components, explaining 79.9% of the total variance (Figure 2A ; Table S5). Principal component 1 explained 56.3% of the total variance and was mainly composed by, in descending order of loading: TNF-α, IFN-ϒ, IL-1β and IL-6 (Table S6). Principal component 2 accounted for 23.6% of the total variance and was mainly composed by IL-6, IL-1β, TNF-α and IFN-γ. We compared labor stages with extracted scores and identified that PC1 plus PC2 scores better distinguish each stage (Figure 2B ) than an independent pro-inflammatory cytokine.
Linear mixed-effects models were fitted by age, multiparity, and number of previous PTBs to examine the effect of labor stages and their interaction with gestational age (week of sample collection) on pro-inflammatory cytokine concentrations. We hypothesized that throughout pregnancy and during stages of labor, women who delivered preterm might have had higher cytokine concentrations levels than those who delivered at term; therefore, we considered subcategories of delivery as a random effect within the model.
We identified positive linear effects of labor stages on concentrations of IFN-γ, TNF-α and IL-1β (Figure 3A-C; Table S7), and positive nonlinear effects on concentrations of IL-6 (Figure 3D; Table S7). However, the variance explained by random effect was null for TNF-α, IL-1β and IL-6 model, and 27.21% for IFN-γ model (Table S7).
When we assessed the moderating variable effect, we found no correlation between gestational age (sample collection week) and IFN-γ, TNF-α and IL-1β concentrations (Figure 3F-H ). In contrast to IL-6 concentrations we found a nonlinear relationship with gestational age (Figure 3I ), in addition to statistically significant stage*sample collection interaction (Table S7), suggesting there is an estimated value for each stage of the relationship between sample collection (gestational age) and IL-6 concentration. Consequently, the relationship between labor stages and IL-6 concentration could be independent from gestational age.
Furthermore, we fitted a model with the pro-inflammatory profile, and found a nonlinear relationship between labor stages and principal-component scores (Figure 3E ; Table S7). As with IL-6 model, we found that this correlation is independent from gestational age (Figure 3J ; Table S7). In addition, we identified that mean scores of term, late preterm and very preterm groups were similar, but far lower compared to extremely preterm group (variance explained by random effect was 86.90%; Table S7).
We previously identified that IL-6 was significantly correlated with changes in labor stages, showing one of the highest average effects on cytokine concentration for a one- unit increase in stages of labor. Therefore, we calculated an optimal cut-point to test the diagnostic performance of IL-6 for the identification of spontaneous labor (Stage 4, effective and regular uterine contractions, and dilation >3 cm) regardless of gestational age (Table 3 ).
The identified cutoff was 34.60 pg/mL and the area under the ROC curve (AUC) was 0.862 (95% CI 0.758-0.967). It is worth noting the negative predictive value: 97.93% (95% CI 92.11-98.71), which indicates that 97.93% of pregnant women are not in established labor with an IL-6 concentration <34.60 pg/mL (truly negative screening test).
To corroborate the assumptions for using an IL-6 cut-point, we assumed homogeneity of outcome on either side of the cut-point (a discontinuous relationship) when X axis is a time variable.14Our time variable is the stage, and the discontinuous relationship we wish to achieve is with respect to Stage 4 and all other stages, thus confirming the validity of dichotomization in this setting (Figure 1C ).
Subsequently, we calculated sampling-to-delivery intervals (censoring those exceeding 120 days) to estimate the probability that a pregnant woman did not give birth according to the estimated cut-point (log-rank test: P =0.0011, Figure 4A ), and fitted a Cox proportional-hazards model by age, multiparity, and number of previous PTBs to examine the association between IL-6 concentrations and the development of delivery from any gestational age. A nonlinear effect was identified indicating a relationship between IL-6 concentrations and increased risk of delivery (HR 202.09, 95% CI 24.57-1662.49, P <0.001; HR [second-degree] 109.74, 95% IC 14.12-853.07,P <0.001; Table S8). Simulated adjusted hazard ratio estimates showed that hazard ratios exponentially increase in according to the identified cut-point (log scale: 3.54 pg/mL, Figure 4B ).
Once the time-dependent adjustment has been applied and we ensured that all women included for analysis developed spontaneous labor leading to delivery, we evaluated the time-varying diagnostic performance of IL-6 with time-dependent ROC curves. Event status is observed at each time point yielding different sensitivity and specificity values throughout follow-up (intervals between sampling and delivery). At each time pointt , each individual is classified as a case or control. A case is defined as any individual experiencing the event (laboring woman) between baseline t =0 and time t and a control as an individual who remains event-free (non-laboring woman) at time t .
Using these definitions, we identified at 12 days an AUC of 0.785 (95%CI 0.693-0.877, Table 4 ), where the diagnostic/predictive performance decreases as time increases. As in previous analyses, the strengths of this marker (at 12 days) are the specificity (81.0%, SE±3.94) and the negative predictive value (84.38%, SE±3.72).