The Mediating Role of Inflammation in the Association Between NHHR and Pelvic Inflammatory Disease: A Case-Control Study Based on Machine Learning

Background

Pelvic inflammatory disease (PID) is a common bacterial infection of the female reproductive system that can cause serious complications, including chronic pelvic pain, infertility, and fallopian tube damage.^1,2 Research shows that around 10–15% of women may develop PID in their lifetime, with young women at the highest risk, especially those with multiple sexual partners, no protection, or a history of sexually transmitted infections (STIs).³ PID is more common in developing countries, and without early treatment, it can lead to long-term complications that significantly impact female fertility and quality of life.^4,5 Recently, the role of metabolic diseases, especially hyperlipidemia, on reproductive health has become a focus.⁶ Hyperlipidemia is not only a major risk factor for heart disease but may also influence the occurrence and progression of PID through disruptions in lipid metabolism and inflammation.^7,8 Research suggests that in hyperlipidemia, fat tissue releases large amounts of pro-inflammatory factors such as tumor necrosis factor-α (TNF-α) and interleukin-6 (IL-6), leading to long-term low-grade inflammation. This chronic inflammation may weaken the immune system, alter the reproductive system’s environment, and increase the susceptibility of pelvic tissues to infections, thereby contributing to the progression of PID.^7,9 Although PID is usually caused by bacterial infections such as Chlamydia trachomatis or Neisseria gonorrhoeae, recent studies¹⁰ suggest that host lipid metabolism may regulate the severity and persistence of infections. Cholesterol can promote pathogen invasion and intracellular survival, which may exacerbate infection related pelvic inflammatory reactions.¹⁰

NHHR, the ratio of non-high-density lipoprotein cholesterol to high-density lipoprotein cholesterol, is a novel lipid marker that reflects lipid metabolism disorders and has been closely associated with the risk of chronic diseases such as cardiovascular disease and diabetes.¹¹ Compared to traditional lipid indices such as TG/HDL or LDL/HDL, NHHR provides a more comprehensive assessment of atherogenic lipid burden and inflammatory potential. By encompassing all cholesterol particles except HDL, it integrates the contributions of LDL, VLDL, and remnants, which are often overlooked by single-lipid markers. High NHHR is closely associated with chronic low-grade inflammation, suggesting a potential role in reproductive system diseases.¹² HDL and LDL are not only lipid transport carriers, but also have immunomodulatory functions. HDL can inhibit monocyte activation and downregulate the expression of inflammatory factors, while oxidized LDL can activate immune cells, promote macrophage aggregation, and amplify local tissue inflammation. These immunomodulatory effects may affect the pelvic immune microenvironment, thereby contributing to susceptibility to PID. Studies⁷ indicate that elevated NHHR may trigger a stronger inflammatory response, which could be a key factor in PID development. Inflammation serves as a crucial link between hyperlipidemia and PID, as chronic low-grade inflammation caused by lipid metabolism disorders can alter immune system function and disrupt the pelvic environment, increasing susceptibility to infections.⁹ Pro-inflammatory factors released from fat tissue, such as C-reactive protein (CRP), IL-6, and TNF-α, not only affect overall immune balance but may also exacerbate PID by activating immune cells, leading to more severe tissue damage.^10,13,14 Moreover, prolonged inflammation can alter the composition of pelvic microbiota, making infections more persistent and further accelerating PID progression.⁷ Therefore, exploring the relationship between NHHR, hyperlipidemia-related inflammatory responses, and PID is crucial for early disease identification, prevention, and precision intervention.

It is hypothesized that NHHR contributes to PID risk through inflammatory pathways, with specific inflammatory markers mediating this relationship. While previous studies have explored the role of hyperlipidemia in systemic inflammation, the direct association between NHHR and PID, as well as the mediating role of inflammatory markers, remains largely unexamined. This study, based on a large-scale clinical dataset, employs advanced machine learning methods combined with multivariate regression, logistic regression, and pathway analysis to systematically investigate the complex relationship between NHHR, hyperlipidemia-related inflammatory responses, and PID, and to evaluate the mediating role of inflammation. Through mediation effect analysis, this study not only reveals the potential mechanisms of hyperlipidemia in PID occurrence but also provides data support and a theoretical foundation for PID risk prediction and intervention strategies.

Methods

Study Population and Design

This case-control study retrospectively collected data from 317 patients who were newly diagnosed with PID at Affiliated Hospital of Jining Medical University between January 2018 and December 2024, along with 3,257 control individuals who were hospitalized during the same period but were confirmed to be free of PID (Figure 1). This study was conducted in accordance with the Helsinki Declaration and was approved by the Medical Science Research Ethics Committee of Jining Medical College Affiliated Hospital (Approval Number: 2025–04-C033).

Figure 1 Research population admission and exclusion process diagram.

The inclusion criteria were as follows:

1. Female patients aged 18 years or older.
2. For the PID group: newly diagnosed PID with surgical or clinical confirmation during hospitalization.
3. For the control group: hospitalized during the same period, confirmed to be free of PID and without gynecologic inflammatory conditions.

The exclusion criteria were as follows:

a) Age below 18 years.

b) Absence of surgical confirmation for PID diagnosis.

c) Presence of malignant tumors.

d) Continuous use of antibiotics or lipid-regulating medications in the past three months.

e) Lack of clinical data and laboratory test results.

After applying these exclusion criteria, a total of 258 newly diagnosed PID patients and 1,989 control participants were included in the final analysis.

Variables

The primary outcome variable in this study was the diagnosis of PID. The intraoperative diagnostic criteria for PID included:

a) Significant congestion on the surface of the fallopian tubes.

b) Edema of the fallopian tube walls.

c) Purulent exudate at the fimbrial end or serosal surface of the fallopian tubes.

PID was defined as infection or inflammation involving the upper female reproductive tract, including the fallopian tubes, ovaries, uterus, and adjacent pelvic structures, confirmed through intraoperative findings.

Baseline characteristics such as age, height, weight, body mass index (BMI), overweight status, past medical history (history of heart disease, hypertension, diabetes, hyperlipidemia, endometriosis, and adenomyosis), surgical history (history of cesarean section, laparoscopic surgery, open surgery, hysteroscopic surgery), and menstrual characteristics (age at menarche, menstrual regularity, menstrual flow, dysmenorrhea) were collected from patients’ medical records by trained professionals upon hospital admission. Laboratory variables included WBC (white blood cell count), PLT (platelet count), LC (lymphocyte count), NC (neutrophil count), MC (monocyte count), PLR (platelet-to-lymphocyte ratio), NLR (neutrophil-to-lymphocyte ratio), MLR (monocyte-to-lymphocyte ratio), SII (systemic inflammation index), HDL (high-density lipoprotein), LDL (low-density lipoprotein), VLDL (very-low-density lipoprotein), TG (triglycerides), TC (total cholesterol), and NHHR (non-HDL cholesterol to HDL cholesterol ratio). Blood samples were taken while fasting, before any tests or treatments, and were analyzed in the lab within one hour of collection.

Calculation formulas for related indices:

BMI= Weight(kg)/Height(cm)²

PLR= PLT(10⁹/L)/LC(10⁹/L)

NLR= NC(10⁹/L)/LC(10⁹/L)

MLR= MC(10⁹/L)/LC(10⁹/L)

SII= PLT(10⁹/L)* NC(10⁹/L)/LC(10⁹/L)

NHHR= no-HDL(mmol/L)/HDL(mmol/L)

Statistical Analysis

Continuous variables were expressed as Mean ± SD, while categorical variables were presented as proportions with 95% confidence intervals (95% CI). To explore the relationship between various factors and the risk of PID, both univariate and multivariate regression analyses were performed. Results were reported as regression coefficient Beta (β), P-value (P), odds ratio (OR), and 95% CI. Subgroup analyses were conducted to identify potential associations within specific populations. Mediation analysis was applied to assess the mediating role of inflammatory indicators (WBC, PLT, LC, NC, MC, PLR, NLR, MLR, SII) in the association between NHHR and PID, with adjustments for relevant covariates. Mediation analysis was conducted using the “mediation” package in R. The significance of the indirect effect was tested using nonparametric bootstrapping with 5,000 resamples to generate 95% confidence intervals (CI) for the average causal mediation effect (ACME). A mediation effect was considered statistically significant if the 95% CI did not include zero. The analysis adjusted for relevant covariates. A mediation effect was considered present if all of the following criteria were met: a significant indirect effect, a significant total effect, and a positive proportion of the mediation effect.¹⁵ The dataset was randomly divided into a training cohort (70%) and a validation cohort (30%). Least absolute shrinkage and selection operator (LASSO) regression was applied as a linear regression technique that utilizes a lambda penalty coefficient to retain variables with non-zero regression coefficients while eliminating those with zero coefficients, ensuring the selection of the most relevant variables for the study.^16,17 In the training cohort, LASSO regression combined with multivariate regression analysis was used to identify key predictors for model construction. A nomogram was then developed to provide a visual representation of the predictive model. Model performance was evaluated through receiver operating characteristic (ROC) curve analysis, calibration curves, and decision curve analysis (DCA). The validation cohort was subsequently used to assess the model’s generalizability. All statistical analyses were conducted using R software (http://www.R-project.org), with P < 0.05 considered statistically significant.

Results

Baseline Characteristics, Univariate, and Multivariate Analysis

Table 1 presents the baseline characteristics of study participants, categorized by NHHR tertiles. The overall NHHR value among all participants was 2.22 ± 0.76, with the low tertile at 1.50 ± 0.24, the middle tertile at 2.12 ± 0.17, and the high tertile at 3.03 ± 0.67. Compared to the low NHHR group, individuals in the high NHHR group were older, had higher BMI, were more likely to be overweight, had no history of cesarean section, and had a higher proportion of menstrual abnormalities. Furthermore, those with high NHHR had higher lymphocyte counts, lower PLR, NLR, and SII values, as well as lower HDL and higher LDL, VLDL, TG, and TC levels (all P < 0.05). Supplementary Tables 1 and 2 present the univariate and multivariate analyses of inflammatory and lipid indices associated with PID. In Model I, NHHR showed a positive association with PID (OR = 1.33, 95% CI = 1.14–1.55). This association remained significant in Model II after adjusting for age and BMI (OR = 1.34, 95% CI = 1.14–1.57) and in Model III after further adjusting for surgical and medical history (OR = 1.38, 95% CI = 1.17–1.64). Even after adjusting for all covariates in Model IV, NHHR remained a significant risk factor for PID (OR = 1.39, 95% CI = 1.16–1.66). Supplementary Table 2 and Supplementary Figure 1 illustrate subgroup analyses, indicating that NHHR is a risk factor for PID among individuals who were overweight, had endometriosis (EM), had or did not have adenomyosis, hypertension, had no history of heart disease, diabetes, hyperlipidemia, had no history of cesarean section, laparoscopic surgery, hysteroscopy, had a history of open surgery, had regular menstrual cycles, moderate menstrual flow, and had dysmenorrhea.

Table 1 Baseline Characteristics Description of the Study Population

Association Between Inflammation, NHHR, and PID & Mediation Analysis

As shown in Table 2, the inflammatory markers WBC, PLT, NC, MC, PLR, NLR, MLR, and SII were positively associated with PID in both unadjusted and adjusted models, with all associations reaching statistical significance (P < 0.05). Table 3 presents the association between NHHR and inflammatory markers after multivariate regression analysis. After adjusting for all factors, NHHR was positively correlated with WBC (β=0.26, 95% CI=0.15–0.37, P < 0.001), LC (β=0.12, 95% CI=0.08–0.16, P < 0.001), NC (β=0.12, 95% CI=0.02–0.21, P = 0.0132), MC (β=0.02, 95% CI=0.01–0.03, P < 0.001), and negatively correlated with PLR (β=−9.69, 95% CI=−13.48–5.89, P < 0.001). NHHR was negatively correlated with PLR (β = −9.69, 95% CI = −13.48 to −5.89, P < 0.001). Mediation analysis (Figure 2) demonstrated that WBC mediated 19.26% of the association between NHHR and PID. In this mediation model, NHHR was the independent variable, WBC acted as the mediator, and PID was the dependent variable. The results indicated a significant indirect effect of NHHR on PID prevalence through WBC, with an indirect effect size of 0.01 (95% CI = 0.00–0.01, P = 0.016) and a direct effect size of 0.02 (95% CI = 0.01–0.04, P = 0.002), suggesting that NHHR influences PID through both direct and indirect pathways, with WBC mediating approximately 19.26% of this relationship. Additionally, the mediating effects of other inflammatory markers, including PLT, LC, NC, MC, PLR, NLR, MLR, and SII, were evaluated, as shown in Supplementary Table 3.

Table 2 Multi Factor Analysis of PID

Table 3 Multivariate Analysis of NHHR and Inflammatory Markers

Figure 2 Analysis of the correlation between leukocyte mediated NHHR and PID.

Construction of a PID Diagnostic Model Based on Lipid and Inflammatory Markers

The study population was randomly divided into training (70%) and validation (30%) cohorts. The training cohort included 1,575 participants, among whom 177 (11.24%) had PID, while the validation cohort included 672 participants, of whom 81 (12.05%) had PID. As shown in Supplementary Table 4, no significant differences were found between the training and validation cohorts in terms of age, BMI, medical history, surgical history, inflammatory markers, or lipid parameters (P > 0.05). LASSO regression was first used to select diagnostic factors for PID (Figure 3A), with 10-fold cross-validation applied for feature selection and normalization (Figure 3B). The optimal lambda (λ) value was selected based on the minimum mean squared error. At λ = 1, 15 predictive variables were selected via LASSO regression combined with multivariate logistic regression analysis, including BMI, endometriosis, adenomyosis, hypertension, diabetes, hyperlipidemia, history of open surgery, PLT, NC, PLR, NLR, HDL, VLDL, TG, and NHHR (Supplementary Table 5). These 15 independent risk factors were included in a multivariate logistic regression model to build a PID prediction model (Supplementary Table 6). A nomogram was created to visually display the diagnostic model (Figure 3C). ROC curve analysis showed that the model had good accuracy, with an area under the curve (AUC) of 0.825 in the training cohort (Figure 3D) and 0.819 in the validation cohort (Figure 3G), confirming its predictive ability. Calibration curve analysis showed a strong match between predicted and actual PID diagnoses in both cohorts (Figures 3E and H). Decision curve analysis (DCA) further confirmed the model’s clinical usefulness in both cohorts (Figures 3F and I).

Figure 3 PID prediction model. (A) shows the variation characteristics of the coefficient of variation. Each curve in the graph represents the coefficient variation of each variable. The vertical axis represents coefficient values, the lower horizontal axis represents log (λ), and the upper horizontal axis represents the number of non-zero coefficients in the model at this time. (B) shows the process of selecting the optimal value of parameter λ in the LASSO regression model using the ten fold cross validation method. (C) is a column chart showing the prediction of concurrent CAL in Kawasaki disease patients based on clinical symptoms and laboratory test results. When using a column chart, use a ruler to draw a vertical line between the target variable and the dot scale at the top of the chart to determine the contribution of each variable to the total score. Add up the number of points for each variable, and then draw a vertical line from the total score at the bottom of the bar chart to the disease outcome to determine the estimated result. (D) shows the ROC curve of the training set based on a column chart, with an AUC value of 0.825. (G) shows the ROC curve of the validation set based on a column chart, with an AUC value of 0.819. (E and H) show the calibration curves of the training set and internal validation set based on column charts, respectively. The dashed line represents the ideal reference line, where the predicted probability matches the observed survival rate, while the solid line is used to calculate the performance of the bar chart. The closer the solid line is to the dashed line, the more accurate the model prediction will be. (F) shows the DCA curve of the training set based on a column chart. (I) shows the DCA curve drawn based on a column chart for internal validation.

Discussion

The findings of this study indicate that higher NHHR is significantly associated with an increased risk of PID. This positive correlation remains robust after adjusting for age, BMI, surgical history, and other factors, with OR values ranging from 1.33 to 1.39. This suggests that elevated NHHR reflects lipid metabolism disorders, which may contribute to PID development through inflammatory responses or immune dysregulation. Inflammatory markers like WBC, PLT, and NLR are closely linked to PID occurrence and strongly associated with high NHHR levels, indicating that inflammation is crucial in PID’s pathogenesis. Mediation analysis revealed that WBC plays a significant mediating role in the relationship between NHHR and PID, accounting for approximately 19.26% of the association. Other inflammatory markers like PLR and NLR also play similar roles to an extent. Machine learning analysis using LASSO regression and multivariate logistic regression models identified BMI, hyperlipidemia, and inflammatory markers as key predictors. The model demonstrated strong predictive performance, achieving AUC values of 0.825 in the training cohort and 0.819 in the validation cohort. These results suggest that machine learning can combine multiple risk factors to provide early prediction and screening tools for PID, improving the clinical use of inflammatory markers in PID diagnosis.

Hyperlipidemia, particularly high NHHR levels, may contribute to PID development by causing chronic low-grade inflammation.^9,12 Compared to traditional lipid indices such as TG/HDL or LDL/HDL, NHHR provides a more comprehensive reflection of atherogenic lipid burden and inflammatory potential. It encompasses all cholesterol particles except HDL, including LDL, VLDL, and remnant lipoproteins, which are often overlooked in single-lipid markers. Emerging evidence suggests that NHHR outperforms TG/HDL and LDL/HDL in predicting cardiovascular risk and systemic inflammation, and may better capture lipid-associated immune alterations in chronic disease contexts.¹¹ In the setting of PID, which may involve subtle immune-metabolic imbalances, NHHR may thus serve as a more integrative and sensitive marker. Disorders in lipid metabolism not only trigger systemic inflammatory responses but also play a key role in the local immune environment. For instance, cytokines and chemokines released by adipocytes can affect the local immune response in the reproductive tract, increasing susceptibility to infection and promoting PID development.^13,18 In hyperlipidemia and metabolic imbalance, inflammation in fat tissue and immune system dysfunction can weaken the body’s ability to fight pelvic infections, increasing the risk of PID.¹⁹ This is further supported by a meta-analysis involving over 168,000 individuals, which showed that patients with metabolic syndrome have significantly elevated total white blood cells, neutrophils, lymphocytes, basophils, and monocytes compared to healthy controls,²⁰ highlighting the systemic inflammatory burden associated with metabolic disturbances. While PID is primarily caused by ascending infections, especially sexually transmitted pathogens, growing evidence suggests that host factors—such as immune dysregulation and chronic low-grade inflammation associated with hyperlipidemia—may influence disease susceptibility, persistence, and clinical outcomes. Studies also show that hyperlipidemia worsens inflammation by damaging blood vessels, making them more permeable, and increasing oxidative stress. These changes can harm blood flow and affect local immune responses.^21,22 These factors may help bacteria settle and spread in the pelvic area, especially in people with high-fat diets or obesity, where lipid metabolism problems make PID even more complex.²³ Therefore, preventing and managing PID should focus not just on treating infections but also on improving lipid metabolism and reducing related inflammation.

Inflammatory markers like WBC and NLR play a key role in the development and progression of PID. Studies²⁴ show that these markers are not just signs of PID but may also be involved in its underlying disease process. Their role in PID is closely linked to the immune system’s response to infection, especially in cases of chronic low-grade inflammation.^25,26 In hyperlipidemia, high inflammatory markers may contribute to PID by triggering local immune responses, worsening tissue damage, and making infections in the reproductive system more severe. Research^27,28 has found a strong link between WBC levels and PID, showing that WBC serves as a systemic inflammation marker, with higher counts often indicating more severe infections. During PID progression, inflammatory markers interact with the local immune response, possibly activating immune cells like neutrophils and monocytes. This can lead to more tissue damage and a stronger inflammatory reaction. Chronic inflammation in PID not only increases immune cell activity, leading to further tissue damage, but also changes the local pH, oxidative balance, and cytokine levels. These changes create a cycle that makes it harder to clear infections and prolongs the disease.^24,29 Because of this, inflammatory markers may not only contribute to PID but also make symptoms worse, especially in people with hyperlipidemia and metabolic disorders. Controlling these inflammatory factors, particularly in those with high lipid levels, may help slow down PID progression and reduce damage. Lowering lipid levels to reduce chronic inflammation or using targeted anti-inflammatory treatments could help relieve PID symptoms and improve immune balance, ultimately leading to better patient outcomes.

This study underscores the potential of machine learning for early PID diagnosis and risk prediction. By combining clinical data and biomarkers, machine learning models have successfully identified key risk factors for PID. Using LASSO regression, the model selected the most relevant predictive variables and created a personalized risk assessment tool.^16,17 The high AUC values and calibration curve analysis confirm the accuracy and clinical value of this model for diagnosing PID. Future improvements could include adding more biomarkers and clinical data to make the model even better for early detection and personalized treatment. Overall, machine learning offers a promising approach to improving PID diagnosis, risk evaluation, and management. It can support clinical decision-making and help achieve better patient outcomes.

Advantages and Limitations

This retrospective case-control study included a relatively large total sample size. However, the number of confirmed PID cases was limited, which may affect the generalizability of the results. The results provide additional evidence supporting the positive correlation between NHHR and PID prevalence. Additionally, the study highlights NHHR as an easily accessible lipid-related indicator for identifying individuals at risk of PID and underscores the mediating role of inflammation in this association. Despite its strengths, this study has several limitations. First, as a retrospective study conducted at a single medical institution in China, there is a risk of recall and reporting bias, and potential selection bias due to reliance on surgically confirmed PID cases. These factors may limit the generalizability of the findings to broader populations. Second, although LASSO regression was used for variable selection, the final model was built using logistic regression alone. Advanced machine learning methods such as decision trees or gradient boosting, which are capable of capturing nonlinear relationships, were not applied due to limitations in resources and technical expertise. Future studies should explore multi-model comparisons to improve prediction performance. Third, although the model demonstrated good discrimination in both training and validation cohorts, the sample size used in the machine learning analysis was relatively limited. To minimize overfitting and enhance internal robustness, we applied LASSO regression and a standard 7:3 split for model evaluation. Nonetheless, external validation in larger, independent, and preferably multi-center cohorts remains necessary to confirm the model’s generalizability. Finally, moderation effects were not explored, and although multiple covariates were adjusted for, residual confounding cannot be fully excluded. These limitations should be carefully addressed in future studies to improve the robustness and interpretability of the findings.

Conclusion

This study suggests that elevated NHHR levels are associated with a higher risk of PID, and this association may be partially explained by inflammatory responses, particularly WBC levels. Although the observed mediating effect of WBC was modest, these findings indicate that NHHR could serve as a useful and accessible marker for identifying individuals at increased risk of PID, especially among those with metabolic abnormalities or gynecological comorbidities. Incorporating NHHR into PID risk assessment may help improve early detection and stratified prevention strategies. Future studies should validate these findings in external cohorts and examine whether NHHR interacts with other clinical or metabolic factors to influence PID susceptibility. Additionally, prospective data and a broader range of inflammatory mediators are needed to better understand the lipid-inflammation-PID pathway.