In total, 20 healthy, nonpregnant female participants aged 56-89 years at Prisma Health Greenville Memorial Hospital with a history of surgical intervention to correct POP via a procedure that used polypropylene mesh were selected for the study. The participants, who were not matched, included 10 who experienced postsurgical mesh complication in which the implanted mesh protruded through the vaginal wall (also referred to as mesh exposure) and 10 participants who did not experience this complication post surgery. This sample size was estimated as an effective cohort for the pilot study using a 1-tailed t test based on an a priori power analysis, which indicates the number of patients for a given theoretical minimum study power as a function of the expected difference between patients with and those without mesh exposure, or the Cohen d. We assumed a conservative study power of 0.80 and a 100%, or 2-fold, difference in the level of a given cytokine between individuals with and without mesh exposure, equivalent to a Cohen d of 1. Here, 20 patients with equal distribution among the 2 groups are needed to observe the difference with a probability of .1. Participants with POP recurrence or taking medications that would alter inflammatory response were excluded from this study.

The study protocol was approved by the institutional review board (IRB) of Prisma Health (Pro00067964). Informed consent from all study participants was obtained using an IRB-approved informed consent form. All samples collected and data analyzed were deidentified and followed IRB protocol.

Blood samples were obtained from the 20 selected participants. Approximately 12 mL of blood was drawn from the upper extremity of each participant into 3 BD Vacutainer EDTA-coated tubes. Deidentified blood samples were then transferred on ice to a laboratory facility at the University of South Carolina School of Medicine Greenville for immediate processing. Each participant’s blood sample was divided into equal aliquots for 24-hour incubation at 37 °C under 3 distinct conditions: (1) incubation with inflammatory agent lipopolysaccharide (LPS) at 20 ng/mL (positive control), (2) incubation with sterile polypropylene mesh area of 2 cm × 2 cm (experimental), and (3) incubation alone (negative control). After incubation, the plasma layer was collected following centrifugation (1500 × g, 10 min, 4 °C) and immediately stored at –80 °C.

Cytokine levels in each blood sample were quantified using the bead-based MILLIPLEX Human Cytokine/Chemokine/Growth Factor Panel A—Immunology Multiplex Assay (EMD Millipore Corp), which is composed of analytes for target cytokines interleukin 1α (IL-1α), IL-1β, IL‑2, IL-4, IL-6, IL-8, IL-10, IL-12p40, IL-12p70, IL-17A, interferon-gamma (IFN-γ, tumor necrosis factor-alpha (TNF-α), and granulocyte-macrophage colony-stimulating factor (GM-CSF). Frozen plasma samples were thawed at room temperature and analyzed following Milliplex protocol guidelines. Cytokine concentrations were measured using a Bio-Plex 200 (Bio-Rad) and Bio-Plex Manager software (Bio-Rad). Sample volume was doubled to ensure measurable levels of cytokines, and assay output data were adjusted to reflect concentrations in plasma samples. Each multiplex assay was performed in duplicate, and cytokine levels were evaluated in 3 independent measurements.

To identify significant associations among the cytokines, PCA was used to examine a total of 60 blood samples (20 participants × 3 blood treatments). Among the 20 participants, 10 experienced postsurgical mesh exposure through the vaginal wall and 10 did not. Figure 1 depicts a biplot of each blood treatment and summarizes the intercorrelated relationships among individual inflammatory mediators. The combined variances explained for PC 1 and PC 2 in blood samples incubated with LPS (Figure 1A), polypropylene mesh (Figure 1B), or alone (Figure 1C) were approximately 64%, 73%, and 66%, respectively. In all 3 treatment groups, IL-10 and IL-4 align in the same directions, as do IL-12p70 and IFN-γ, indicating a positive correlation for both of these cytokine pairs. In contrast, IL-6 and IL-12p40 were negatively correlated when comparing stimulation of blood via LPS (Figure 1A) versus polypropylene mesh (Figure 1B). Only IL-1α displayed a negative correlation when comparing blood incubated with polypropylene mesh (Figure 1B) versus blood incubated alone (Figure 1C).

When PCA was used to examine only blood samples incubated with polypropylene mesh, PC 1 and PC 2 explained 60.1% and 13.1% of the total data variance, respectively (Figure 1B). Figure 2A displays each cytokine’s contribution to PC 1 and illustrates that IFN-γ, IL-12p70, and IL-2 are the predominant contributors to the variance explained in PC 1. In addition, IL-1α, IL-17A, and TNF-α exhibited contributions above a level expected if the contributions were uniform. All other cytokines have contributions to PC 1 similar to or less than what would be expected if the contributions of all cytokines were uniform. Figure 2B illustrates that the predominant contributors to the variance explained in PC 2 are IL-10, IL-4, and IL-6. All other cytokines have contributions to PC 2 similar to or less than what would be expected if the contribution of all cytokines were uniform.

In order to visualize associations between the participants presenting the absence or presence of postsurgical mesh exposure through the vaginal wall, a biplot illustrating individual participants was created (Figure 3). This biplot reveals a high percentage of variability represented by the first 2 PCs (79.1%). Blood samples from participants who did not experience postsurgical mesh exposure were heavily represented by positive PC 1 values, while blood samples from participants with the presence of postsurgical mesh exposure were generally represented by positive PC 2 values.

Four supervised machine learning models incorporating all 13 cytokines were trained using 70% (42/60) of the available 60 observations (20 participants × 3 independent measurements); the remaining 30% (18/60) was used to test the models’ accuracy when predicting the presence of mesh exposure through the vaginal wall. All 4 machine learning machines achieved at least 62% (26/42) training accuracy (Table 1). Artificial neural network achieved the highest prediction accuracy of 83% (15/18), while decision tree and Naïve Bayes both achieved a prediction accuracy of 61% (11/18). Naïve Bayes, decision tree, and artificial neural network excelled at correctly predicting patients with the presence of mesh exposure postsurgery at 89% (16/18). Artificial neural network was superior for correctly predicting patients who did not experience mesh exposure postsurgery (14/18, 78%).

Additional models and predictive analyses explored whether a smaller set of cytokines could achieve similar predictive results. Feature selection using a random forest method identified a group of 7 cytokines capable of yielding effective predictive analysis: IL-1β, IL-8, IL-12p40, IL-12p70, TNF-α, IL-17A, and IL-6. Figure 4 illustrates that models exhibited variation among the importance of cytokines when implementing this more targeted group of cytokines. IL-1 and IL-8 are strongly represented in all models, while IL-6 is important in only Naïve Bayes.

Table 2 illustrates that all models achieved at least 64% (27/42) training accuracy. The logistic regression model that used the 7 selected cytokines achieved a training accuracy of 81% (34/42), a prediction accuracy of 72% (13/18), a sensitivity of 67% (12/18), and a specificity of 78% (14/18), and thus outperformed compared to the logistic regression model that incorporated all of the cytokine data. Moreover, decision tree models achieved the same result when using the selected cytokines or when all cytokines were included. The prediction accuracy in Naïve Bayes and artificial neural network models executed with the 7 selected cytokines decreased by only 5% each compared to the same models that used all of the cytokine data.

Among patients with POP who undergo mesh implantation surgery, 17% of them experience mesh exposure through the vaginal wall [18]. This rate of surgical mesh complication is significant when compared to 0.035%-5.4% mesh-related erosions reported in other mesh-based surgeries [19-23], necessitating the development of a personalized decision support tool for patients with POP. This exploratory study demonstrates a novel and efficient approach to predicting postsurgical outcomes for mesh implantation using cytokine levels in patient blood following exposure to a biomaterial. Previous studies have often used patient demographic and medical data to train machine learning programs to create predictive outcomes for POP mesh surgeries [13]. In contrast, this study uses biological material to mimic an in vivo response, thus presenting a novel, noninvasive, personalized clinical decision tool. A systematic PCA approach identifies associations among cytokines that provide physiological insight. Supervised machine learning models developed in this study demonstrate that blood cytokine measurements may be used as a predictive tool. In addition, the number of cytokine measurements needed may be reduced without compromising predictive capabilities, rendering this approach more applicable within a clinical setting.

The PCA analysis illustrated in Figure 1 reveals several significant associations among the cytokines. Several cytokines display positive correlations when comparing the 2 different stimuli: LPS and polypropylene mesh. However, IL-6 and IL-12p40 are negatively correlated between these 2 treatments. Thus, these 2 cytokines may explain the different inflammatory responses induced by LPS versus polypropylene mesh. When comparing blood samples incubated alone to those incubated in the presence of polypropylene mesh, only IL-1α exhibits a negative correlation, demonstrating that this proinflammatory mediator might be specifically affected by the mesh stimulus. Furthermore, 2 pairs of cytokines positively correlate (IL-10 and IL-4; IL-12p70 and IFN-γ), indicating that one of the cytokines in each pair could be eliminated to reduce the number of cytokines tested in a clinical setting.

The cytokines that contribute most to each PC segregate into proinflammatory and anti-inflammatory cytokines (Figure 2). When cytokine data from patient blood incubated with mesh were analyzed using PCA, cytokines IFN-γ, IL-12p70, and IL-2 were the largest contributors to the variance explained in PC 1. These markers are identified as proinflammatory agents [24-26], which suggests that proinflammatory cytokines may heavily influence PC 1. In contrast, cytokines IL-10, IL-4, and IL-6 were the largest contributors to the variance explained in PC 2. IL-4 and IL-10 are prominent anti-inflammatory cytokines [25], suggesting that anti-inflammatory cytokines heavily influence PC 2. IL-6, previously thought to have proinflammatory function only, is recently recognized as potentially having both proinflammatory and anti-inflammatory roles in COVID-19 [27] and diabetes [28].

When juxtaposing the biplot of polypropylene-stimulated cytokine observations (Figure 1B) with that of mesh exposure outcome (Figure 3), it can be extrapolated that proinflammatory cytokines IL-12p40, IL-1α, and TNF-α are positioned in the region of the biplot that uniquely corresponds to surgical outcomes involving the presence of mesh exposure through the vaginal wall. Such juxtaposition suggests that IL-12p40, IL-1α, and TNF-α may be associated with the presence of postsurgical mesh exposure. These observations may inspire potential therapeutic strategies that could improve postsurgical outcome. For example, the surgical mesh could be designed to modulate these key proinflammatory cytokines. In this way, while supporting the pelvic structure, the mesh could simultaneously function in controlling the cytokine response to minimize biomaterial rejection.

Table 1 describes the accuracy of supervised machine learning models and demonstrates that cytokines can exhibit predictive capabilities. Previous studies have performed predictive analysis for POP using risk factors derived from patient medical history [13]. However, such data can be incomplete and inaccurate [29]. This study demonstrates the utility that measured responses of biological samples can also have in developing robust predictive models. Chen et al [12] similarly used blood cytokine levels in a machine learning study to predict severe AKI after cardiac surgery. Their study concluded that a logistic regression model was the most effective in discovering the cytokine associations in severe AKI. In this study, the prediction accuracy for all 4 models exceeded 60%, with the artificial neural network model demonstrating the best overall performance, predicting POP surgical outcomes with 83% (15/18) accuracy when trained with all 13 cytokines. This predictive capability is similar to that reported for prediction derived from patient medical history [4], despite this study comprising a significantly smaller patient group. Considering the small population size, these results represent relatively high prediction accuracy for health care data.

When creating models trained with a subgroup of 7 cytokines (Table 2), selected using a random forest method, the artificial neural network model maintained the greatest effectiveness with respect to sensitivity, specificity, and prediction accuracy. Moreover, the group of selected cytokines outperformed the larger group of cytokines in the logistic regression model and achieved the same results in the decision tree model. The Naïve Bayes and artificial neural network prediction accuracy dropped only 5% when using the subgroup of cytokines, thus demonstrating the resiliency of these models. These results demonstrate that predictive capabilities are retained with fewer cytokines, which would enhance clinical feasibility by reducing the cost and time associated with this clinical decision tool.

This study implemented rigorous research methods to identify physiological relationships among cytokine markers and developed robust machine learning models to predict mesh exposure; yet, some limitations should be noted. First, because this is a pilot study, the sample size is limited to 20 participants within a single hospital system. Nevertheless, this limited sample size predicted 83% (15/18) accuracy, a level that compares favorably with another predictive model study by Chu et al [9] that achieved a prediction accuracy of 71% in a study population size of 467. Thus, the results of this pilot study indicate the utility of this approach and the merit of future studies. Future study will provide validation with a larger population of participants from multiple hospitals. Additionally, the 10 participants in each group were not matched regarding variables. To minimize confounders, patients with POP recurrence or taking medication that would alter inflammatory response were excluded and the age ranges and average age at the time of surgery within each group were similar. Future studies with a larger population, however, will benefit from matching participants with respect to these and other potentially confounding variables. Nonetheless, the results of this pilot study highlight the importance of inflammatory markers in the prediction of this postsurgical condition.

While this preliminary study is limited to a sample size of just 20 participants, this novel approach to using cytokine response to predict POP surgical outcomes has successfully distinguished important cytokines and their correlations. Moreover, these relationships point toward prospective therapies that could promote better surgical outcomes. Supervised learning models also demonstrate a high level of accuracy, specificity, and sensitivity, even when a smaller group of cytokine data is used. This result suggests that blood cytokine analysis might be feasibly used in a clinical setting to predict POP surgical outcomes. Further study with a larger patient population will be needed to confirm the utility of this method. If successful at a larger scale, this approach has the potential to change perspectives in which surgeons would recommend and proceed with POP repair surgeries and to prevent undesired outcomes of mesh-related surgeries in patients.