Blood Count-Derived Inflammatory Indices at Two Timepoints in Insulin-Treated Gestational Diabetes: A Retrospective Cohort Study

Abstract

Purpose

Complete blood count-derived inflammatory indices are widely studied in pregnancy, with inconsistent findings; if they capture a persistent maternal characteristic, a woman’s value at one timepoint should locate her value at another. We examined within-woman tracking of six indices, and their relation to fetal overgrowth, in insulin-treated gestational diabetes (GDM).

Patients and Methods

Retrospective cohort of 358 women with insulin-treated GDM at a tertiary centre. Six indices (NLR, PLR, MLR, SII, SIRI, PIV) were calculated at the oral glucose tolerance test (OGTT, 24–28 weeks) and at delivery admission, two physiologically distinct occasions. The primary outcome was large-for-gestational-age (LGA) birth (INTERGROWTH-21st); macrosomia was secondary. Tracking was quantified by Spearman correlation and the between-woman variance share from a mixed model. Analytic samples were 358 (group comparisons, discrimination), 353 (paired) and 284 (adjusted models), one lower in each for MLR, SIRI and PIV.

Results

LGA occurred in 139 infants (38.8%). No index discriminated LGA to a clinically useful degree (AUC 0.51–0.57). Adjusted for body-mass index (BMI) and HbA1c, SIRI was associated with LGA (OR 1.64 per standard deviation, 95% CI 1.18–2.39; n = 283) and survived Bonferroni correction (corrected P = 0.040), though not consistently across samples; we regard this nominal finding as exploratory. BMI and HbA1c, per standard deviation, had odds ratios of 1.42 and 1.48 and numerically higher AUCs (0.63 each). All six indices changed between the two occasions (all P<0.001); within-woman rank correlation was weak to moderate (ρ 0.27–0.54) and the between-woman variance share 0.25–0.56.

Conclusion

CBC-derived indices tracked poorly within women between the two occasions, and none discriminated LGA to a clinically useful degree. Whether this reflects instability of the underlying inflammatory state or the differing clinical circumstances of the two samples cannot be determined from this design.

Introduction

Gestational diabetes mellitus is one of the most common metabolic disorders of pregnancy, defined as glucose intolerance first recognized during gestation, and it affects a large proportion of pregnancies worldwide.^1–3 It increases the risk of adverse maternal and perinatal outcomes, particularly large-for-gestational-age birth, macrosomia, and caesarean delivery.^3,4 Its pathogenesis is now understood to centre on chronic low-grade systemic inflammation and immune dysregulation.^2–5 Activated immune cells and the sustained release of pro-inflammatory cytokines feed the insulin resistance that characterizes the condition.^3,5,6

Complete blood count-derived inflammatory indices, including the neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), monocyte-to-lymphocyte ratio (MLR), systemic immune-inflammation index (SII), systemic inflammation response index (SIRI), and pan-immune-inflammation value (PIV), are inexpensive, widely available markers of systemic inflammation and have established prognostic value in oncology and metabolic disease.^5,7–10 They have been evaluated across many pregnancy complications, but the findings are inconsistent. SII, SIRI, and PIV predicted preeclampsia in some cohorts,¹¹ whereas NLR, PLR, and SII lost independent significance for preterm birth after adjustment for metabolic factors,¹² several studies of NLR in GDM found no association,⁵ and a recent prospective study reported only limited discriminatory performance of NLR, PLR and MLR for the early prediction of GDM.¹³ One reason may be that a single measurement cannot capture a changing inflammatory state; studies have typically reported one value, taken at the oral glucose tolerance test, without following how it changes across gestation.⁵ Studies using more than one sample have compared trimesters and have focused on the development of GDM rather than on birth outcomes,^14,15 and a recent longitudinal study followed six indices at seven points from early pregnancy to the postpartum period in women with diet-controlled GDM.¹⁶ What has not been examined is how these indices behave between a mid-pregnancy sample and one taken at delivery in an insulin-treated population, and how they relate to fetal overgrowth in that population.

Two points need to be separated at the outset. The first is whether a woman’s index value at one point in pregnancy locates her value at another: if these indices capture a persistent maternal characteristic, the ranking of women should be preserved between occasions even though the values themselves shift with gestation. The second is that the two occasions studied here, the diagnostic oral glucose tolerance test at 24–28 weeks and admission for delivery, are physiologically distinct states, so that a change between them combines physiological progression, the peripartum context and measurement variability, and the design cannot apportion these. We therefore refer to within-woman tracking rather than to reproducibility, which would imply repeated measurement under comparable conditions. Considering the role of inflammation in gestational diabetes mellitus, we measured six complete blood count-derived inflammatory indices at these two timepoints in women with insulin-treated GDM, examined within-woman tracking between them, and related the measurements to large-for-gestational-age birth and macrosomia.

Materials and Methods

Study Design and Setting

We conducted a retrospective cohort study at the Department of Perinatology, Başakşehir Çam ve Sakura City Hospital, a tertiary referral centre in Istanbul, Türkiye. The study was reported in accordance with the STROBE guidelines for observational research;¹⁷ the completed checklist, with page numbers, is provided as Supplementary Material. Because the centre houses a level III neonatal intensive care unit and receives regional referrals for threatened preterm labour and other complications, the delivering population has a high proportion of preterm and complicated pregnancies, a feature we return to when interpreting our findings.

Participants

We identified all women with insulin-treated gestational diabetes mellitus (GDM) admitted to our hospital between 1 May 2020 and 10 March 2026; the window opens with the opening of the centre, which admitted its first obstetric patients on 1 May 2020. Eligible patients were those whose discharge summary contained both “GDM” and “insulin”, those for whom an insulin prescription report had been issued during pregnancy, or those with an ICD code of O24.4 (gestational diabetes mellitus) who were receiving insulin. GDM had been diagnosed by a one-step 75-g OGTT at 24–28 weeks using the IADPSG criteria (fasting ≥92, 1-h ≥180, or 2-h ≥153 mg/dL), a single abnormal value being sufficient for diagnosis;¹ insulin had been initiated when glycaemic targets were not met with medical nutrition therapy alone (fasting capillary glucose below 95 mg/dL and either 1-hour postprandial glucose below 140 mg/dL or 2-hour postprandial glucose below 120 mg/dL)¹⁸ and was titrated to the same targets. Where a woman had more than one hospital admission, the records were merged so that each patient entered the analysis once.

The diagnostic criteria in use at the centre, and the thresholds for initiating and titrating insulin, did not change over the study period. SARS-CoV-2 infection and vaccination status were not systematically recorded and could not be retrieved; they are an unmeasured source of variability in the cell counts across a recruitment window that opened during the pandemic, and we acknowledge them as such rather than attempting to reconstruct them.

From 977 insulin-treated GDM patients, we applied the exclusions summarised in Figure 1. Women were excluded for pre-eclampsia, pregestational (type 1 or type 2) diabetes, active documented infection, multiple pregnancy, intrahepatic cholestasis of pregnancy, chronic renal or hepatic disease, intrauterine fetal death, or termination, and for having no recorded HbA1c, an incomplete blood count, or an unavailable delivery record. These categories overlap, as many patients met more than one criterion; the final cohort comprised 358 patients; the population available for each analysis is given in Figure 1.

Figure 1.

Study flow diagram. Of 977 women identified with insulin-treated gestational diabetes mellitus, those with missing data or with comorbid or pregnancy conditions were excluded, leaving an analysis cohort of 358. Exclusion categories overlap because many women met more than one criterion, so the listed numbers do not sum to the difference between 977 and 358. The lower box gives the population available for each analysis; it is one woman fewer for the three monocyte-containing indices, because one woman had no valid monocyte count at the OGTT.

Abbreviations: GDM, gestational diabetes mellitus; ICD, International Classification of Diseases; OGTT, oral glucose tolerance test; HbA1c, glycated haemoglobin; LGA, large-for-gestational-age; BMI, body-mass index; MLR, monocyte-to-lymphocyte ratio; SIRI, systemic inflammation response index; PIV, pan-immune-inflammation value.

Two of these exclusions require comment, because the conditions concerned arise after the exposure was measured. Pre-eclampsia (n = 47) and intrahepatic cholestasis of pregnancy (n = 12) were ascertained from the delivery record rather than at the 24–28 week timepoint, so excluding these women conditions the sample on a post-baseline variable and may introduce selection bias. They were excluded because both conditions are themselves accompanied by systemic inflammatory change, which would have confounded the comparison of index values between the two occasions. Retaining these women within a sensitivity analysis would have been preferable, but the differential white-cell counts needed to compute the indices were not part of the dataset extracted for this study, so that analysis could not be performed. We regard this as a limitation rather than a resolved point.

The use of HbA1c as an eligibility criterion likewise requires justification, since HbA1c is neither an exposure nor an outcome but an adjustment covariate. It entered the eligibility criteria because a contemporaneous measure of glycaemic severity was considered necessary for the cohort to be interpretable, but the conventional approach is to define the cohort on exposure and outcome availability and to address covariate missingness within the analysis. The analyses described below as full-cohort analyses are therefore conditioned on an HbA1c value having been recorded at some point in pregnancy. Within that cohort, we repeated the analyses in all 358 women irrespective of whether an HbA1c value fell within the OGTT window, as described under Statistical analysis.

We excluded pregestational diabetes deliberately. Long-standing diabetes may be accompanied by vascular disease that can produce either fetal overgrowth or, through uteroplacental insufficiency, fetal growth restriction.^19,20 Including these pregnancies would have introduced a group in which the relationship between metabolism and birth weight is not unidirectional, obscuring the question we set out to study. Restricting the cohort to insulin-treated but non-pregestational GDM therefore yielded a more homogeneous population, at the cost (acknowledged in the Discussion) of removing the women with the highest insulin requirements and greatest macrosomia risk.

Maternal Characteristics

Body-mass index was calculated from weight and height measured at the first antenatal visit to our centre. The gestational age at that visit varied between women, some presenting in the first and others in the second trimester, and it was not recoverable from the records; the value is therefore a measured booking body-mass index rather than a pre-pregnancy body-mass index, and is not directly interchangeable with the pre-pregnancy values used in much of the literature. Maternal weight at the time of insulin dosing was not recorded, so insulin dose could not be normalised to body weight.

HbA1c was taken as the value measured between 24 weeks 0 days and 28 weeks 6 days of gestation, the window within which the diagnostic OGTT is routinely performed in our unit; the gestational week recorded with each measurement was used to decide this, and the values within the window fell at a median of 25 weeks. Where a woman had more than one HbA1c during pregnancy, only a value falling within this window was carried forward. The dataset does not hold the date of either test, so the interval between the HbA1c measurement and the OGTT cannot be expressed in days; both fall within the same 24 to 28 week window, which bounds the separation at about five weeks and is as precise as the records allow. Values outside the window were not used: for 24 women the only value available was earlier than 24 weeks and for 50 it was later than 28 weeks, so 74 women had no HbA1c within the window and were excluded from the adjusted models on that basis. In nine women the OGTT result or the HbA1c value was retrieved from the national electronic health record or from an emergency department note rather than from the hospital laboratory system.

Blood-Count Indices and Timepoints

Complete blood counts were retrieved at two predefined timepoints: at the diagnostic OGTT (24–28 weeks) and at the peripartum admission, the latter defined as the count taken on admission for delivery, before birth. Where more than one count was available at a timepoint, we used the one closest to the diagnostic OGTT at the mid-pregnancy timepoint and the first count recorded on admission at delivery. The records available to us do not carry the laboratory dates of these tests: we therefore cannot report the proportion of mid-pregnancy counts drawn on the same day as the OGTT or the interval between the two, and whether each count was drawn in the fasting state is not recorded. All counts were measured on a Sysmex XN-series haematology analyser (Sysmex Corporation, Kobe, Japan) in the hospital’s central laboratory, and both the OGTT and delivery-admission samples were analysed there. Neutrophil, lymphocyte, monocyte and platelet counts are expressed as ×10⁹/L. From the absolute neutrophil, lymphocyte, monocyte, and platelet counts we calculated six indices: the neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), monocyte-to-lymphocyte ratio (MLR), the systemic immune-inflammation index (SII = platelet × neutrophil/lymphocyte;),⁷ the systemic inflammation response index (SIRI = neutrophil × monocyte/lymphocyte;),⁸ and the pan-immune-inflammation value (PIV = neutrophil × platelet × monocyte/lymphocyte;).⁹ Counts falling outside prespecified plausibility limits (neutrophils 0.5–25, lymphocytes 0.3–8, monocytes 0.05–3 and platelets 50–800 ×10⁹/L) were treated as missing rather than carried into the derived indices. One recorded OGTT monocyte value fell outside these limits and was treated as missing, so the three monocyte-containing indices (MLR, SIRI and PIV) were computed in one fewer woman at that timepoint.

A delivery-admission count could not be retrieved for five of the 358 women, who were therefore excluded from the paired analyses (n = 353). Because all six indices are derived from a single count, missingness did not differ across indices; the same five women were absent from every paired comparison.

The two timepoints were chosen to contrast two different states: the OGTT count reflects the mid-pregnancy metabolic state, well before delivery, whereas the admission count is taken at presentation for delivery. These are physiologically distinct occasions rather than repeated measurements under comparable conditions, and they were separated by a median of 11.7 weeks. Sampling the same women twice allowed us to examine how far a woman’s value on one occasion locates her value on the other, that is, within-woman tracking, rather than reproducibility in the metrological sense. Comparisons involving mode of delivery were exploratory, since labour status at the time of sampling was not recorded; nor were cervical dilatation, induction, rupture of membranes, or exposure to corticosteroids or antibiotics before sampling systematically documented.

Outcomes

The primary outcome was large-for-gestational-age (LGA) birth, defined as a birth weight above the 90^th centile for gestational age and sex using the INTERGROWTH-21^st newborn standard,²¹ computed with the gigs package in R.²² Gestational age was assigned from the last menstrual period when it agreed with the first-trimester crown-rump length measurement to within 7 days, from the crown-rump length when the two differed by more than 7 days, and from the last menstrual period alone when no first-trimester scan was available. Gestational age was recorded as completed weeks plus days and supplied to the centile calculation in that form, without rounding to whole weeks. Infant sex was supplied to the standard for every infant, so sex-specific centiles were used throughout; Table 1 reports sex as a raw count only. Birth weight was measured in the delivery room immediately after birth on calibrated electronic scales.

Table 1.

Baseline Characteristics of the Analysis Cohort (n = 358)

Characteristic	Value
Age, years	33.0 (29.0–37.0)
Gravidity	3.0 (2.0–4.0)
Parity	1.0 (1.0–2.0)
Body-mass index at the first antenatal visit, kg/m2	32.3 (28.8–36.1)
HbA1c at the OGTT, % (n = 284)	5.7 (5.4–6.1)
Gestational age at delivery, weeks	37.9 (37.0–38.3)
Birth weight, g	3335 (2996–3780)
Total daily insulin dose, IU/day (n = 358)	15.0 (8.0–28.0)
Basal insulin (detemir), IU/day, in the 311 women receiving basal insulin	8.0 (6.0–14.0)
Bolus insulin (aspart), IU/day, in the 236 women receiving bolus insulin	12.5 (8.0–24.0)
Insulin regimen, n (%)
Basal and bolus	189 (52.8)
Basal only	122 (34.1)
Bolus only	47 (13.1)
Infant sex, n (male/female)	181/177
Mode of delivery, n (%)
Caesarean	297 (83.0)
Vaginal	61 (17.0)
Large-for-gestational-age, n (%)	139 (38.8)
Macrosomia (birth weight ≥ 4000 g), n (%)	57 (15.9)
Preterm birth (<37 weeks), n (%)	77 (21.5)
Birth before 33 weeks 0 days, n (%)	17 (4.7)

Notes: Data are median (interquartile range) unless otherwise indicated. Body-mass index was calculated from weight and height measured at the first antenatal visit. HbA1c, glycated haemoglobin measured within the window of the diagnostic oral glucose tolerance test (OGTT, 24 to 28 weeks); a value was available in 284 of 358 women and all other characteristics are based on the full cohort. The basal and bolus doses are given for the women who received that component; the total daily dose is the sum of the two components in all 358 women. Large-for-gestational-age was defined as birth weight above the 90^th centile for gestational age and sex by the INTERGROWTH-21^st standards. No infant weighed exactly 4000 g.

Abbreviation: IU, international units.

Seventeen infants were born before 33 weeks 0 days and therefore fall outside the range of the INTERGROWTH-21^st Newborn Size Standard, which covers 33 weeks 0 days to 42 weeks 6 days. The gigs package applies the INTERGROWTH-21^st Very Preterm Size at Birth reference to these infants automatically, and this is the instrument that was used for them. A sensitivity analysis excluding these seventeen infants is reported in the Results.

We chose LGA rather than absolute macrosomia as the primary outcome because a fixed weight threshold does not account for gestational age and may miss relative overgrowth in preterm infants; in this cohort a fifth of infants were born preterm. Macrosomia, retained as a secondary outcome so that the two definitions could be compared, was defined as a birth weight of 4000 g or more. No infant weighed exactly 4000 g, so this definition is equivalent to a birth weight above 4000 g.

The INTERGROWTH-21^st standard, derived from a healthy low-risk reference population, is known to yield higher LGA rates than national or customised charts;^23–25 the LGA rate observed in this cohort is considered in the Discussion.

Statistical Analysis

Continuous variables are summarised as median (interquartile range) and compared with the Mann–Whitney U-test; the distribution-free approach was chosen because the indices are right-skewed. Categorical variables are reported as counts and percentages.

Analytic samples differed across analyses and are stated with each result: 358 women for the unadjusted group comparisons and for discrimination, 353 for the paired two-timepoint analyses, and 284 for the models adjusted for both body-mass index and HbA1c; each of these is one fewer for the three monocyte-containing indices.

Each index was examined separately and never entered together in a single model. The indices are mathematically interrelated, sharing neutrophil, lymphocyte, monocyte, and platelet terms, so combining them would introduce severe collinearity; we therefore fitted a separate logistic regression for each index, each adjusted for body-mass index (BMI) and HbA1c at the OGTT. Odds ratios are expressed per one standard deviation, so that effect sizes are comparable across indices measured on different scales; the standard deviations used for scaling were computed within the analytic sample of each model and are reported with the corresponding table. Discrimination was assessed by the area under the receiver-operating-characteristic curve, with its 95% confidence interval computed by the DeLong method, and areas under the curve were compared between predictors by the DeLong test rather than by inspection. Collinearity in the adjusted models was checked with variance inflation factors, which did not exceed 1.07.

As a further check on whether the indices reflected treatment intensity, we examined the correlation between total daily insulin dose and each index using the Spearman rank coefficient, alongside the correlation of dose with HbA1c and BMI for comparison. Total daily dose was taken as the last recorded dose before delivery and calculated as the sum of basal (detemir) and bolus (aspart) units per day; the regimen was classified from these recorded doses. Because this dose is recorded near delivery rather than at the OGTT, the correlation was also examined using the delivery-admission indices.

To examine how far a woman’s index value on one occasion locates her value on the other, we compared the OGTT and delivery-admission values with the Wilcoxon signed-rank test and quantified within-woman rank tracking with the Spearman coefficient, whose confidence interval was obtained by Fisher’s z transformation with the Bonett–Wright standard error. Because a rank correlation measures association rather than agreement, we supplemented it with a variance-component analysis. Each index was log-transformed and a linear mixed model was fitted with a random intercept for woman and a fixed effect for timepoint, from which the between-woman and within-woman variance components were read directly. Because the timepoint enters as a fixed effect, the systematic shift between occasions is removed by construction rather than being absorbed into the error term, so the resulting intraclass correlation coefficient is of the consistency form. An absolute-agreement coefficient was not used: the two occasions are prespecified and physiologically distinct rather than randomly sampled replicates, and an absolute-agreement coefficient would have been driven largely by the gestational change that the Wilcoxon tests already establish. Confidence intervals for the variance partition were obtained by bootstrap resampling of women, with 1000 replicates and seed 20,260,830.

The within-woman median change and its confidence interval were estimated by the Hodges–Lehmann method, computed as the median of the Walsh averages of the paired differences, with a percentile bootstrap confidence interval based on 2000 replicates and the same seed. To test whether the change between timepoints differed by mode of delivery, we compared the within-woman change (delivery admission minus OGTT) between women who delivered vaginally and those delivered by caesarean, with Bonferroni correction across the six indices; this comparison is reported as exploratory and is presented in the Supplementary Material.

We did not dichotomise the indices, search for optimal cut-offs, or impute missing values; all analyses were complete-case. This was a deliberate stance: searching for thresholds across many index–outcome combinations inflates false-positive findings, and an honest account of the indices’ behaviour is better served by treating them as continuous predictors.

Where a nominal association emerged, we examined whether it was stable rather than reporting it at face value. As sensitivity analyses we refitted the adjusted models after log-transforming the indices and after excluding observations with a Cook’s distance above 4/n. The latter is an influence diagnostic rather than a validation: it indicates how far an estimate depends on a small number of observations, and estimates obtained after such exclusions should not be read as more reliable than the estimates from the full sample. Bonferroni correction was applied within each family of six indices, not across the study as a whole, and this was prespecified. Approximately fifty hypothesis tests were performed in total; the correction therefore controls the family-wise error rate within each analysis, not the study-wide rate, and the corrected P values should be read with that in mind.

Because the single index finding that survived correction arose within the subset of women with an HbA1c value in the OGTT window, and because that subset was defined by an eligibility criterion rather than by the analysis itself, we repeated the analyses in ways that separate selection from modelling. We report, for all six indices: the unadjusted group comparison and discrimination within the 284 women with an HbA1c value as well as in the full cohort of 358; an unadjusted logistic model in the full cohort of 358; and a logistic model adjusted for body-mass index alone, which requires no HbA1c value and can therefore be fitted in all 358 women. Throughout, “unadjusted group comparison” denotes the Mann–Whitney test and “unadjusted logistic model” the logistic model containing the index alone, and the analytic n is given with each.

Each primary model included three predictors: one index, BMI, and HbA1c. Adjustment was limited to these two variables because maternal adiposity and glycaemia are the established metabolic determinants of fetal overgrowth, rather than on the basis of a fixed events-per-variable rule. We recognise that HbA1c is not a complete proxy for glycaemic severity and that other determinants of birth weight were available in the record; we therefore prespecified an extended sensitivity model adding maternal age and parity to BMI and HbA1c, and a further model adding the fasting glucose value of the OGTT in the women for whom it was recorded. Gestational weight gain, a history of previous macrosomia or LGA, and glycaemic control in later pregnancy were not recorded and could not be included. All analyses were performed in R version 4.6.0 (R Foundation for Statistical Computing, Vienna, Austria), using the gigs package for birth weight centiles, pROC for receiver-operating-characteristic analysis and lme4 for the mixed models. Two-sided P values below 0.05 were considered statistically significant.

Ethics

The study was approved by the Scientific Research Ethics Committee No. 1 of Başakşehir Çam ve Sakura City Hospital on 25 February 2026 (protocol no. 2026–64, decision no. 64) and was conducted on anonymised routine clinical records in accordance with the Declaration of Helsinki.

Results

Cohort Characteristics

The 358 women had a median age of 33 years and a median BMI of 32.3 kg/m², placing half the cohort in the obese range (Table 1). Median HbA1c at the diagnostic OGTT was 5.7% in the 284 women with a value within the window. The median total daily insulin dose was 15 IU, and 189 women (53%) were treated with a basal–bolus regimen. Delivery was predominantly by caesarean (297 of 358, 83%), and 77 women (21.5%) delivered preterm, 17 of them before 33 weeks 0 days. The date of recruitment is not held in the study dataset and could not be recovered, so we report the year of delivery instead, reconstructed from the recorded last menstrual period and the gestational age at delivery. It could be assigned for 279 of the 358 women and was distributed as follows: 9 in 2020, 44 in 2021, 31 in 2022, 49 in 2023, 62 in 2024, 59 in 2025 and 25 in 2026; for the remaining 79 women no reliable date was available.

By the INTERGROWTH-21^st standard, 139 infants (38.8%) were large-for-gestational-age and 57 (15.9%) were macrosomic by absolute birth weight.

Inflammatory Indices and LGA

At the OGTT timepoint (24–28 weeks), none of the six indices separated LGA from non-LGA pregnancies to a degree that would be clinically useful (Table 2 and Figure 2). In unadjusted group comparisons in the full cohort (n = 358 for SII, NLR and PLR; 357 for MLR, SIRI and PIV), median values were similar between groups (Figure 2A), and the areas under the ROC curve ranged from 0.51 to 0.57, with confidence intervals including or close to 0.50 (Figure 2B). On the Mann–Whitney test PIV reached nominal significance (P = 0.035) and SII approached it (P = 0.052). BMI (AUC 0.63, 95% CI 0.57–0.68, n = 358) and HbA1c (AUC 0.63, 0.57–0.70, n = 284) had numerically higher AUCs than the indices. By the DeLong test the AUC of BMI exceeded those of PLR and MLR (P = 0.010 and 0.009) but did not differ significantly from those of SII, NLR, SIRI or PIV (P between 0.08 and 0.18); the BMI and HbA1c AUCs are computed in different samples and are not directly comparable with one another.

Table 2.

Inflammatory Indices at the Oral Glucose Tolerance Test and Fetal Overgrowth: Unadjusted Analyses in the Full Cohort and Adjusted Analyses in the Complete-Case Sample

Variable	n	Mann–Whitney P	AUC (95% CI)	Unadjusted OR (95% CI)	P	Bonf. P	Adjusted OR (95% CI)	P	Bonf. P	ρ with Insulin dose (P)
A. Large-for-gestational-age birth, full cohort (n = 358, 139 events); adjusted model contains body-mass index only
SII	358	0.052	0.56 (0.50–0.62)	1.29 (1.04–1.60)	0.022	0.13	1.26 (1.02–1.58)	0.037	0.22	−0.07 (0.18)
NLR	358	0.14	0.55 (0.49–0.61)	1.20 (0.97–1.50)	0.090	0.54	1.24 (0.99–1.56)	0.067	0.40	−0.08 (0.15)
PLR	358	0.76	0.51 (0.45–0.57)	0.98 (0.79–1.21)	0.84	1.00	0.95 (0.76–1.18)	0.65	1.00	−0.07 (0.16)
MLR	357	0.66	0.51 (0.45–0.58)	1.08 (0.87–1.33)	0.50	1.00	1.10 (0.88–1.37)	0.39	1.00	−0.04 (0.40)
SIRI	357	0.12	0.55 (0.49–0.61)	1.42 (1.10–1.89)	0.012	0.070	1.44 (1.11–1.92)	0.011	0.064	−0.03 (0.61)
PIV	357	0.035	0.57 (0.50–0.63)	1.42 (1.12–1.84)	0.006	0.036	1.39 (1.10–1.80)	0.009	0.056	−0.04 (0.44)
BMI	358	—	0.63 (0.57–0.68)	—	—	—	1.53 (1.22–1.92)	<0.001	—	+0.04 (0.45)
B. Large-for-gestational-age birth, complete-case sample with HbA1c (n = 284, 112 events); adjusted model contains body-mass index and HbA1c
SII	284	0.037	0.57 (0.51–0.64)	1.36 (1.07–1.76)	0.014	0.085	1.30 (1.01–1.70)	0.042	0.25	—
NLR	284	0.046	0.57 (0.50–0.64)	1.29 (1.02–1.67)	0.041	0.25	1.35 (1.04–1.80)	0.029	0.18	—
PLR	284	0.88	0.49 (0.43–0.56)	0.99 (0.77–1.25)	0.92	1.00	0.95 (0.74–1.22)	0.69	1.00	—
MLR	283	0.76	0.49 (0.42–0.56)	1.17 (0.92–1.51)	0.19	1.00	1.22 (0.96–1.58)	0.11	0.67	—
SIRI	283	0.027	0.58 (0.51–0.65)	1.65 (1.20–2.36)	0.004	0.024	1.64 (1.18–2.39)	0.007	0.040	—
PIV	283	0.011	0.59 (0.52–0.66)	1.55 (1.17–2.13)	0.004	0.026	1.44 (1.10–1.97)	0.015	0.092	—
BMI	284	—	0.63 (0.56–0.69)	—	—	—	1.42 (1.10–1.85)	0.007	—	—
HbA1c	284	—	0.63 (0.57–0.70)	—	—	—	1.48 (1.14–1.96)	0.005	—	+0.16 (0.006)
C. Macrosomia, complete-case sample with HbA1c (n = 284, 45 events); adjusted model contains body-mass index and HbA1c
SII	284	—	—	1.51 (1.14–2.01)	0.004	0.026	1.43 (1.06–1.93)	0.018	0.11	—
NLR	284	—	—	1.44 (1.09–1.89)	0.009	0.051	1.47 (1.09–2.03)	0.015	0.089	—
PLR	284	—	—	1.22 (0.90–1.63)	0.19	1.00	1.19 (0.86–1.61)	0.28	1.00	—
MLR	283	—	—	1.21 (0.90–1.59)	0.18	1.00	1.27 (0.93–1.70)	0.10	0.61	—
SIRI	283	—	—	1.30 (0.99–1.80)	0.077	0.46	1.30 (0.99–1.76)	0.055	0.33	—
PIV	283	—	—	1.37 (1.03–1.89)	0.037	0.22	1.32 (0.99–1.77)	0.048	0.29	—
BMI	284	—	—	—	—	—	1.49 (1.09–2.06)	0.014	—	—
HbA1c	284	—	—	—	—	—	1.52 (1.11–2.08)	0.009	—	—

Notes: n is the number of women contributing to each row; it is one fewer for MLR, SIRI and PIV because one woman had no valid monocyte count at the oral glucose tolerance test. The unadjusted group comparison is the Mann–Whitney U-test of LGA against non-LGA pregnancies (group medians are shown in Figure 2A) and was prespecified for LGA only. AUC, area under the receiver-operating-characteristic curve, with its 95% confidence interval by the DeLong method. Odds ratios (OR) are per one-standard-deviation increase in the variable, each index entered in a separate logistic model; the unadjusted logistic model contains the index alone, and the adjusted model additionally contains the covariates named in the panel heading. The BMI and HbA1c estimates in the adjusted columns are taken from the SII model and were materially unchanged across models. Standard deviations used for scaling were, in the full cohort, 481 (SII), 1.73 (NLR), 48.5 (PLR), 0.145 (MLR), 1.72 (SIRI), 468 (PIV) and 5.81 kg/m² (BMI), and in the complete-case sample, 480 (SII), 1.81 (NLR), 47.8 (PLR), 0.153 (MLR), 1.83 (SIRI), 486 (PIV), 5.80 kg/m² (BMI) and 0.647% points (HbA1c). P is the uncorrected P value; Bonferroni-corrected (Bonf.) P values were computed from the unrounded P values multiplied by six, the number of indices tested in each family, capped at 1.00; correction was applied within each family of six indices and not across the study. In the full cohort the AUC of BMI was compared with that of each index by the DeLong test (P: SII 0.13, NLR 0.084, PLR 0.010, MLR 0.009, SIRI 0.10, PIV 0.18). ρ, Spearman rank correlation of the variable with total daily insulin dose, with its uncorrected P value in parentheses, computed in the full cohort (n = 358 for SII, NLR, PLR and BMI, 357 for MLR, SIRI and PIV, and 284 for HbA1c). Variance inflation factors in the adjusted models did not exceed 1.07. Sensitivity analyses of the adjusted models are given in Table S1. —, not computed.

Abbreviations: BMI, body-mass index; HbA1c, glycated haemoglobin; LGA, large-for-gestational-age; NLR, neutrophil-to-lymphocyte ratio; PLR, platelet-to-lymphocyte ratio; MLR, monocyte-to-lymphocyte ratio; SII, systemic immune-inflammation index; SIRI, systemic inflammation response index; PIV, pan-immune-inflammation value.

Figure 2.

Discrimination of the indices for large-for-gestational-age (LGA) birth. (A) Index values at the oral glucose tolerance test in pregnancies with (blue) and without (orange) LGA. Boxes show the median and interquartile range, whiskers extend to 1.5 times the interquartile range, and values beyond are plotted individually; y axes differ between panels. (B) Receiver-operating-characteristic curves for each index and for body-mass index (BMI), with the area under the curve given in the key; the diagonal line indicates no discrimination. All curves are computed in the full cohort (n = 358 for SII, NLR, PLR and BMI; 357 for MLR, SIRI and PIV). HbA1c is not shown because its curve would be computed in the different analytic sample of the 284 women with an HbA1c value within the OGTT window; its area under the curve is given in Table 2.

Abbreviations: SII, systemic immune-inflammation index; NLR, neutrophil-to-lymphocyte ratio; PLR, platelet-to-lymphocyte ratio; MLR, monocyte-to-lymphocyte ratio; SIRI, systemic inflammation response index; PIV, pan-immune-inflammation value.

Adjusted Associations with LGA

In logistic models adjusted for BMI and HbA1c, BMI was associated with LGA (OR 1.42 per standard deviation, 95% CI 1.10–1.85, P = 0.007) and so was HbA1c (OR 1.48, 1.14–1.96, P = 0.005); both remained associated with macrosomia (Table 2 and Figure 3). These models were fitted in the 284 women with an HbA1c value within the OGTT window (283 for the three monocyte-containing indices), with 112 LGA events and 45 macrosomia events; 74 women were not included because no HbA1c value fell within the window. Variance inflation factors did not exceed 1.07.

Figure 3.

Adjusted odds ratios for large-for-gestational-age (LGA) birth and for macrosomia. Each point is the odds ratio per one-standard-deviation increase with its 95% confidence interval, from a separate logistic model containing that index together with body-mass index (BMI) and glycated haemoglobin (HbA1c); the BMI and HbA1c estimates shown are from the SII model. Inflammatory indices are shown in blue and the metabolic variables BMI and HbA1c in orange. The dashed vertical line marks an odds ratio of 1 and the x axis is logarithmic. Models were fitted in 284 women with 112 LGA events and 45 macrosomia events (283 women and 111 LGA events for MLR, SIRI and PIV).

Abbreviations: SII, systemic immune-inflammation index; NLR, neutrophil-to-lymphocyte ratio; PLR, platelet-to-lymphocyte ratio; MLR, monocyte-to-lymphocyte ratio; SIRI, systemic inflammation response index; PIV, pan-immune-inflammation value.

Among the indices, SIRI showed the strongest association with LGA in this sample, both in an unadjusted logistic model (OR 1.65, 1.20–2.36, P = 0.004; n = 283) and after adjustment (OR 1.64, 1.18–2.39, P = 0.007), and was the only index to remain significant after Bonferroni correction (corrected P = 0.040). PIV behaved similarly (unadjusted OR 1.55, 1.17–2.13; adjusted OR 1.44, 1.10–1.97, P = 0.015) but did not survive correction (corrected P = 0.092). SII and NLR reached nominal significance in the adjusted models (P = 0.042 and 0.029), and PLR and MLR showed no association.

We examined how far the SIRI finding depended on the analytic sample and on the modelling scale, because it arose within a subset defined by HbA1c availability and because SIRI is strongly right-skewed (Table 2). In the full cohort, the unadjusted group comparison for SIRI gave P = 0.115 and an AUC of 0.55 (n = 357), whereas within the HbA1c subset the same comparison gave P = 0.027 and an AUC of 0.58 (n = 283); the association was therefore somewhat stronger in the subset. An unadjusted logistic model in the full cohort gave OR 1.42 (1.10–1.89, P = 0.012), and a model adjusted for BMI alone, which requires no HbA1c value and could be fitted in all 357 women, gave OR 1.44 (1.11–1.92, P = 0.011); neither survived Bonferroni correction (corrected P = 0.070 and 0.064). In the prespecified extended model adding maternal age and parity to BMI and HbA1c, the SIRI estimate was similar (OR 1.72, 1.23–2.52, P = 0.004, corrected P = 0.024; n = 282); with the fasting OGTT glucose value added, the sample fell to 179 women and the estimate was 1.56 (1.02–2.49, P = 0.048, corrected P = 0.29). After excluding the 17 infants born before 33 weeks 0 days, the BMI-adjusted estimate was 1.40 (1.08–1.89, P = 0.019, corrected P = 0.115). PIV followed a similar pattern, surviving correction in the unadjusted full-cohort model (corrected P = 0.036) but not in the BMI-adjusted one (corrected P = 0.056). Taken together, SIRI and PIV were nominally significant in every logistic model examined, although the unadjusted group comparison for SIRI was not significant in the full cohort (P = 0.115), and their survival of correction for multiple testing depended on the sample and the specification; we regard them as exploratory.

As sensitivity analyses on the prespecified adjusted model, the SIRI association was of similar size after log transformation (OR 1.42, 1.10–1.85, P = 0.008) and after excluding nine observations with a Cook’s distance above 4/n (OR 1.67, 1.19–2.48, P = 0.008); the latter is an influence diagnostic rather than a validation, and estimates obtained after such exclusions should not be read as more reliable than those from the full sample. Full sensitivity output is given in Table S1.

For macrosomia the pattern was similar but weaker: SII and NLR reached nominal significance in the adjusted models (P = 0.018 and 0.015), PIV lay at the threshold (OR 1.32, 0.99–1.77, P = 0.048) and SIRI just above it (OR 1.30, 0.99–1.76, P = 0.055); no index remained significant after Bonferroni correction. BMI (OR 1.49, P = 0.014) and HbA1c (OR 1.52, P = 0.009) remained associated with macrosomia (Table 2 and Figure 3).

Within-Woman Tracking Between the Two Timepoints

Paired values at both timepoints were available for 353 women (352 for MLR, SIRI and PIV), with a median interval of 11.7 weeks (IQR 10.0–13.3) between the OGTT at 26.0 weeks (24.0–28.0) and delivery admission at 37.9 weeks (37.0–38.3). All six indices changed between the two occasions (all P < 0.001; Table 3): by the Hodges–Lehmann estimate SII rose by a median of 94 (95% CI 42 to 145), NLR by 0.78 (0.56 to 1.01), MLR by 0.09 (0.07 to 0.10), SIRI by 1.33 (1.13 to 1.54) and PIV by 253 (203 to 302), whereas PLR fell by 13 (9 to 17). Within-woman rank correlation between the two occasions was weak to moderate (Figure 4): 0.27 (0.16–0.37) for NLR, 0.30 (0.20–0.40) for SIRI, 0.37 (0.27–0.46) for SII, 0.39 (0.29–0.48) for MLR, 0.41 (0.32–0.50) for PIV and 0.54 (0.46–0.61) for PLR.

Table 3.

Inflammatory Indices at the Two Timepoints and Within-Woman Tracking Between Them

Index	n	At the OGTT, Median (IQR)	At Delivery Admission, Median (IQR)	Within-Woman Change, Median (95% CI)	Wilcoxon P	Spearman ρ (95% CI)	ICC (95% CI)
SII	353	923 (730–1241)	1005 (791–1394)	+93.5 (41.9 to 145.3)	<0.001	0.37 (0.27–0.46)	0.37 (0.26–0.46)
NLR	353	3.88 (3.19–4.69)	4.59 (3.60–5.75)	+0.78 (0.56 to 1.01)	<0.001	0.27 (0.16–0.37)	0.25 (0.13–0.35)
PLR	353	132.9 (103.7–161.1)	113.7 (91.1–143.3)	−12.9 (−17.0 to −8.8)	<0.001	0.54 (0.46–0.61)	0.56 (0.45–0.64)
MLR	352	0.33 (0.27–0.42)	0.42 (0.32–0.54)	+0.086 (0.070 to 0.101)	<0.001	0.39 (0.29–0.48)	0.40 (0.30–0.50)
SIRI	352	2.45 (1.81–3.46)	3.80 (2.64–5.23)	+1.33 (1.13 to 1.54)	<0.001	0.30 (0.20–0.40)	0.29 (0.19–0.38)
PIV	352	594 (432–887)	822 (569–1281)	+252.6 (203.0 to 301.5)	<0.001	0.41 (0.32–0.50)	0.40 (0.30–0.48)

Notes: Values are for the women with a blood count at both timepoints (n = 353; 352 for MLR, SIRI and PIV because one woman had no valid monocyte count at the OGTT). OGTT, oral glucose tolerance test at 24 to 28 weeks; delivery admission, the count taken on admission for delivery, before birth; IQR, interquartile range. The within-woman change is delivery admission minus OGTT, estimated by the Hodges–Lehmann method as the median of the Walsh averages of the paired differences, with a percentile bootstrap 95% confidence interval (CI) from 2000 resamples of women (seed 20260830); P values are from the Wilcoxon signed-rank test. ρ is the Spearman rank correlation between a woman’s two measurements, with a 95% CI from the Fisher z transformation using the Bonett–Wright standard error. ICC is the consistency-form intraclass correlation coefficient, the proportion of variance on the logarithmic scale attributable to differences between women, from a linear mixed model of the log-transformed index with a fixed effect for timepoint and a random intercept for woman; its 95% CI is from a percentile bootstrap of 1000 resamples of women (same seed). The between-woman and within-woman variance components on the log scale were, respectively: SII 0.071 and 0.123; NLR 0.036 and 0.109; PLR 0.068 and 0.054; MLR 0.053 and 0.079; SIRI 0.078 and 0.189; PIV 0.140 and 0.212.

Abbreviations: NLR, neutrophil-to-lymphocyte ratio; PLR, platelet-to-lymphocyte ratio; MLR, monocyte-to-lymphocyte ratio; SII, systemic immune-inflammation index; SIRI, systemic inflammation response index; PIV, pan-immune-inflammation value.

Figure 4.

Within-woman tracking between the two timepoints. Each point is one woman, plotting her index value at the diagnostic oral glucose tolerance test (OGTT, 24–28 weeks) against her value at delivery admission; both axes are logarithmic. The dashed line is the line of identity. ρ is the Spearman correlation between a woman’s two measurements (n = 353; 352 for MLR, SIRI and PIV). Distance from the identity line represents the within-woman difference between the two occasions; rank preservation is quantified by the Spearman coefficient, and the corresponding intraclass correlation coefficients are given in Table 3.

Abbreviations: SII, systemic immune-inflammation index; NLR, neutrophil-to-lymphocyte ratio; PLR, platelet-to-lymphocyte ratio; MLR, monocyte-to-lymphocyte ratio; SIRI, systemic inflammation response index; PIV, pan-immune-inflammation value.

The variance-component analysis gave the same picture. On the logarithmic scale, with the timepoint entered as a fixed effect, the between-woman share of the remaining variance was 0.25 (95% CI 0.13–0.35) for NLR, 0.29 (0.19–0.38) for SIRI, 0.37 (0.26–0.46) for SII, 0.40 (0.30–0.50) for MLR, 0.40 (0.30–0.48) for PIV and 0.56 (0.45–0.64) for PLR (Table 3). For five of the six indices, therefore, more than half of the variation that remains after the systematic shift between occasions is removed lies within women rather than between them; PLR is the exception, with a between-woman share just above one half.

Mode of Delivery

An exploratory comparison of the indices by eventual mode of delivery, at both timepoints and for the within-woman change, is presented in the Supplementary Material (Table S2, Figure S1). Because labour status at the time of the delivery-admission sample was not recorded, these comparisons cannot distinguish an association with the route of delivery from the effect of labour already in progress, and they are reported as descriptive only.

Insulin Dose and Inflammatory Indices

Total daily insulin dose did not correlate with any of the six OGTT indices (all ρ between −0.03 and −0.08, all P ≥ 0.15; Table 2). It did correlate with HbA1c (ρ = 0.16, P = 0.006, n = 284) but not with BMI (ρ = 0.04, P = 0.45, n = 358). Because insulin dose was recorded close to delivery, we repeated the analysis using the delivery-admission indices, which are contemporaneous with it. Correlations remained weak (ρ between −0.07 and −0.11; n = 353); NLR and MLR reached nominal significance (P = 0.044 and 0.046) but neither survived Bonferroni correction.

Discussion

In this retrospective cohort of women with insulin-treated GDM, we examined six CBC-derived inflammatory indices at two timepoints. At the OGTT no index showed clinically useful discrimination for LGA. In the prespecified adjusted model SIRI was associated with LGA and persisted through correction for multiple testing, but this persistence depended on the analytic sample and the model specification; BMI and HbA1c were associated with both LGA and macrosomia. All six indices changed between the OGTT and delivery admission within the same women, and a woman’s value at one occasion tracked her value at the other only weakly to moderately, with between-woman differences accounting for a quarter to a half of the variance that remained once the systematic shift between occasions was removed. Insulin dose correlated with HbA1c but with none of the indices. An exploratory comparison by mode of delivery is reported in the Supplementary Material.

The clearest finding was that a woman’s index at one occasion tracked her value at the other only weakly to moderately. All six indices changed between the OGTT and delivery admission, which is not in itself surprising, since labour and the approach to delivery alter the blood count. What is less expected is that the ranking of women was not preserved: the correlation between a woman’s two measurements was only weak to moderate (ρ 0.27 to 0.54), and the variance-component analysis, which removes the systematic shift between occasions by construction, gave the same answer: for five of the six indices, less than half of the remaining variation lay between women. Had the indices reflected a persistent maternal characteristic, women with higher values in mid-pregnancy should still have had higher values at delivery, whatever the overall shift. PLR also moved in the opposite direction to the other five, so the indices did not change together. Tracking of this order means that a single measurement locates a woman only loosely within the distribution at another point in pregnancy. Whether this reflects instability of the underlying inflammatory state or the differing clinical circumstances of the two samples cannot be determined from this design: the two occasions were separated by a median of 11.7 weeks and drawn in different physiological states, and physiological change across gestation, the peripartum context, and analytical and biological variability may all contribute.

Studies of these indices in pregnancy have given inconsistent results, and our data suggest one reason. At the OGTT the indices behaved unevenly: SIRI and PIV showed the largest odds ratios, SII and NLR reached nominal significance, and PLR and MLR showed nothing. That six closely related measures, derived from the same blood count, should behave so differently is worth noting. The SIRI association survived correction for multiple testing in the prespecified adjusted model and when maternal age and parity were added, but not in the full cohort without the HbA1c restriction, where it fell just short of the corrected threshold, and its size depended on the modelling scale: the per-standard-deviation estimate on the raw scale is sensitive to the upper tail of a right-skewed distribution in a way that a rank test is not, and in the full cohort the rank-based comparison for SIRI was not significant. We regard the finding as exploratory: it arose in a single retrospective cohort, its survival of correction depends on the sample and the specification, and its standalone discrimination was poor (AUC 0.55 to 0.58). Whatever the statistical significance, all six indices showed poor standalone discrimination: every confidence interval for the area under the curve included or approached 0.50.

This pattern fits what is known about these indices. They change across pregnancy: the SII peaks in the second trimester and the SIRI rises throughout gestation,¹⁵ and the NLR reaches its maximum in the second trimester.⁵ In one GDM cohort, most indices that differed from controls in the first trimester no longer did so in the second, with two exceptions, which the authors attributed to physiological haemodilution, expanding plasma volume, and shifts in leukocyte composition.¹⁴ They are also affected by transient factors including stress, intercurrent infection, medication, and the time and season of sampling.^3,26,27 A single measurement reflects only one point in time, and as Pace and Vassallo note, one measurement does not fully capture the inflammatory state in GDM.⁵ In our cohort, the indices differed between the two timepoints in the same women. A single measurement can therefore vary within an individual, which may contribute to the different conclusions reached by studies that sample at different times. These indices are arithmetic combinations of routine cell counts and are proposed as surrogates for systemic inflammation; they are not direct measurements of cytokines, immune-cell phenotype, or the maternal-fetal interface, and limited tracking of the surrogate does not by itself establish instability of the underlying immune state.

A recent longitudinal study of women with diet-controlled gestational diabetes sampled six routine blood-derived inflammatory indices at seven points from early pregnancy to the postpartum period and reported distinct group-by-time trajectories relative to normoglycaemic controls, with only postpartum SIRI and pre-delivery PLR remaining significant after correction for multiple testing.¹⁶ Our design differs in that it followed insulin-treated pregnancies at two clinically defined occasions and related the indices to birth outcomes rather than to a normoglycaemic comparison group, but both studies point in the same direction: these indices shift across gestation, and most single-index associations weaken once multiplicity is taken into account.

The insulin dose data are consistent with this picture but do not extend it: dose correlated with HbA1c but with none of the indices at either timepoint. We do not read this as showing that the indices are independent of the metabolic burden of the disease. Dose was recorded as units per day rather than per kilogram, adherence and the adequacy of titration were not known, and dose in an insulin-treated cohort reflects the clinician’s response to glycaemia as much as the glycaemia itself. What the data show is narrower: within a cohort in which all women required insulin, the amount required did not track the indices. Oluklu et al reported that, among women with GDM, those who required insulin had a higher first-trimester SIRI than those managed with diet, and a broader inflammatory burden across second-trimester indices.¹⁴ Our cohort cannot address that comparison, since all women were insulin-treated.

BMI and HbA1c were associated with LGA and with macrosomia in our cohort, and were the only variables associated with both. This is consistent with well-established evidence: pre-pregnancy BMI, gestational weight gain, and HbA1c are independent predictors of fetal overgrowth in GDM,²⁸ and maternal obesity is associated with increased birth weight.²⁹ The BMI in our study was measured at the first antenatal visit rather than reported for the pre-pregnancy period, and the gestational age at that visit varied between women, so the comparison with the pre-pregnancy BMI literature is indirect. Maternal hyperglycaemia increases glucose transfer across the placenta and stimulates fetal insulin secretion, which drives overgrowth of insulin-sensitive tissues; this is the classical Pedersen hypothesis.^6,30 BMI and HbA1c also had numerically higher areas under the curve than any index, although their values were themselves only 0.63, and on formal comparison the BMI AUC exceeded those of PLR and MLR but did not differ significantly from those of the other four indices.

The closest study to ours is that of Bulu and Bulu, who examined inflammatory indices and macrosomia in GDM at the OGTT timepoint and reported that NLR and NHR independently predicted macrosomia.²⁸ We studied insulin-treated GDM rather than a general GDM population, and a partly different set of indices. We also examined whether the associations we found were stable instead of reporting them at face value: of the four indices reaching nominal significance for LGA, only SIRI persisted after correction for multiple testing in the prespecified model, and its persistence was not consistent across analytic samples; for macrosomia none did. Bulu and Bulu reported their associations without these checks, and noted themselves that no multiple-comparison adjustment or model validation was performed.²⁸ Reported discrimination for these indices also depends on how cohorts are assembled; in one GDM study the authors noted that excluding pregnancies with adverse outcomes may have inflated the discriminative capacity of the indices in their low-risk cohort.¹⁴ An index can therefore reach significance in one cohort and not in another, which is consistent with the limited within-woman tracking we observed between the two sampled occasions.

At delivery admission, several indices were higher among women who subsequently delivered vaginally than among those delivered by caesarean, and the SII difference survived correction for multiple testing (Table S2 and Figure S1). A similar pattern has been reported for total leukocyte counts, with higher admission counts in women who went on to deliver vaginally.²⁶ We present this comparison as descriptive only. Labour status at the time of sampling was not recorded, cervical dilatation is unknown, induction and prelabour rupture of membranes are grouped together, and elective and intrapartum caesarean sections are not distinguished, so the blood count may have been drawn after labour was already established. Labour is an inflammatory process involving systemic leukocytosis, myeloid-cell infiltration at the maternal-fetal interface, and activation of the NLRP3 inflammasome,³¹ and a raised index in a woman who proceeds to deliver vaginally may be a consequence of labour in progress rather than a predictor of the route of delivery. The temporal ordering that would make this an association with the eventual mode of delivery has not been established, and the within-woman change between the two timepoints did not differ significantly between delivery groups.

Our cohort had a high LGA rate of 39%. Three factors are likely to contribute, and the choice of standard is only one of them. The INTERGROWTH-21^st standard gives higher LGA rates than customised or national charts, but in GDM cohorts assessed against it the rate has been around a quarter,^23–25 so the standard alone does not account for the figure observed here. Cohort selection contributes more: our centre is a tertiary referral hospital to which women with complicated pregnancies are referred, and the cohort was further restricted to women whose glycaemia required insulin, so it represents the high-risk end of the GDM spectrum; the high caesarean and preterm rates reflect the same selection. Dating accuracy may also contribute, since a centile-based outcome depends directly on gestational age: dating was by last menstrual period where this agreed with the first-trimester crown-rump length and by crown-rump length otherwise, but a proportion of women presented for the first time in the second trimester, and misdating near the 90^th centile would misclassify infants in either direction. The LGA rate should therefore be read as a property of this population and its ascertainment rather than as a general estimate for insulin-treated GDM.

Pregestational diabetes differs from GDM in an important way. Maternal microangiopathy and vasculopathy are complications of longstanding diabetes,^30,32 whereas GDM is by definition first recognised during pregnancy and has a much shorter duration. In pregestational diabetes, microvascular complications can reduce uteroplacental blood flow and produce uteroplacental malperfusion, restricting fetal growth,^30,32 so hyperglycaemia and vascular disease push birth weight in opposite directions.^19,20 Women with pregestational diabetes and vascular complications are considerably more likely to have growth-restricted infants and less likely to have macrosomic ones.¹⁹ Studies of GDM therefore commonly exclude women with pre-existing diabetes,^2,3,11,28 and restricting our cohort to insulin-treated GDM gave a population in which the metabolic drive toward overgrowth was not offset by vascular growth restriction.

Strengths and Limitations

The main strength of this study is that the indices were measured at two contrasting timepoints in the same women, one reflecting the mid-pregnancy metabolic state and one taken at admission for delivery, which allowed within-woman tracking between the two occasions to be examined directly rather than inferred, both by rank correlation and by a variance-component analysis that removes the systematic shift between occasions. We also examined whether the associations we found were stable instead of reporting them at face value, and the cohort was restricted to insulin-treated GDM, reducing heterogeneity in treatment category.

The study was retrospective and conducted at a single tertiary referral centre, to which women with complicated pregnancies are preferentially referred; the LGA, caesarean and preterm rates show that the cohort is a highly selected, high-risk population, and the findings may not extend to diet-controlled GDM or to unselected populations. The cohort was limited to insulin-treated GDM so that the relationship between insulin dose and the indices could be examined.

Analyses were complete-case. HbA1c entered the eligibility criteria, which is not the conventional approach for an adjustment covariate, and the analyses described as full-cohort are therefore conditioned on an HbA1c value having been recorded at some point in pregnancy; we have addressed this within the available data by repeating the analyses in the full cohort without the OGTT-window restriction, but women in whom no HbA1c was recorded at any point could not be recovered. No HbA1c value within the OGTT window was available in 74 women, so the adjusted models were fitted in 284 rather than 358. LGA prevalence was similar in the two groups (112 of 284 and 27 of 74), and included and excluded women did not differ appreciably in age, BMI or birth weight, but this does not establish that the missingness is ignorable, and no imputation was performed. Pre-eclampsia and intrahepatic cholestasis were ascertained after the exposure had been measured and were applied as exclusions; this conditions the sample on post-baseline events and may introduce selection bias, and the differential counts needed to retain these women in a sensitivity analysis were not available.

Adjustment was limited to BMI and HbA1c in the primary model, with maternal age, parity and fasting OGTT glucose examined in sensitivity models; gestational weight gain, a history of previous macrosomia or LGA, and glycaemic control in later pregnancy were not recorded, and residual confounding from these sources cannot be excluded. BMI was measured at the first antenatal visit at a gestational age that varied between women and could not be recovered. Maternal weight at the time of insulin dosing was not recorded, so dose could not be normalised to body weight. SARS-CoV-2 infection and vaccination status were not recorded and represent an unmeasured source of variability in the cell counts across a recruitment window that opened during the pandemic.

We excluded women with documented active infection, but subclinical or undocumented inflammatory conditions at the time of sampling could not be identified. The delivery-admission sample was taken at a clinically heterogeneous moment: women were admitted at different cervical dilatations, for induction, or for prelabour rupture of membranes, and neither labour status nor exposure to corticosteroids or antibiotics before sampling was recorded; the records also did not distinguish elective from intrapartum caesarean. This is why the comparison by mode of delivery is presented as descriptive and in the Supplementary Material, and the vaginal delivery group was small (n = 57 to 61), which limits the precision of that comparison. Because each index was measured once at each timepoint, we could not separate true within-woman change from the analytical and biological variability of the blood count itself, and the two samples were taken a median of 11.7 weeks apart in different physiological states; whether the limited tracking reflects instability of the underlying inflammatory state or the differing circumstances of the two samples cannot be determined from this design. The laboratory dates of the mid-pregnancy count and the OGTT are not held in the records available to us, so the mid-pregnancy count is defined by its position in the record rather than by a measured interval from the OGTT. One woman lacked a valid monocyte count at the OGTT, so the three monocyte-containing indices were analysed in one fewer woman. The OGTT sample preceded insulin treatment and subsequent care, so later glycaemic control and treatment intensity may lie on the pathway between the biomarker and birth weight. Analyses beyond the primary LGA outcome, including the delivery-admission comparison and the insulin dose correlations, were exploratory.

Prospective multicentre studies in unselected GDM populations are needed to determine whether these observations generalise.

Conclusion

In this cohort of insulin-treated GDM, associations between CBC-derived inflammatory indices and fetal overgrowth were at most modest, and none showed clinically useful standalone discrimination; BMI and HbA1c were the variables associated with both LGA and macrosomia. The indices also tracked poorly within women between mid-pregnancy and delivery admission, so that a single measurement locates a woman only loosely within the distribution at another point in pregnancy. Whether this reflects instability of the underlying inflammatory state or the differing clinical circumstances of the two samples cannot be determined from this design, but either way it may help to explain why studies sampling at different times have reached different conclusions.

Funding Statement

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Data Sharing Statement

The data that support the findings of this study are available from the corresponding author on reasonable request. The data are not publicly available because they contain information that could compromise the privacy of participants.

Ethics Statement

The study was approved by the Scientific Research Ethics Committee No. 1 of Başakşehir Çam ve Sakura City Hospital on 25 February 2026 (protocol no. 2026-64, decision no. 64) and was conducted in accordance with the Declaration of Helsinki. Because the study was retrospective and used anonymised routine clinical records, the requirement for written informed consent was waived by the ethics committee.

Consent for Publication

Not applicable. This study used anonymised routine clinical records and contains no individually identifiable person’s data, images, or details.

Author Contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors report no conflicts of interest in this work.