Three Routine Checkup Measures Screen for Severe Sleep Apnea Without Questionnaires

A clinical prediction model combining biological sex, waist circumference, and the triglyceride-glucose index was developed and pilot-tested to triage individuals at risk for severe obstructive sleep apnea without relying on subjective screening questionnaires. In an initial retrospective training cohort of 271 snoring patients who underwent diagnostic polysomnography at the People’s Hospital of Guangxi Zhuang Autonomous Region, the model demonstrated an area under the receiver operating characteristic curve of 0.751. When evaluated in a pilot temporal validation cohort of 82 patients assessed in 2025 at the same hospital’s Cognitive and Sleep Medicine Center, the model achieved an area under the curve of 0.648. At an optimized low risk cutoff of 0.238, the tool demonstrated a sensitivity of 87.2 percent, a negative predictive value of 68.4 percent, and a specificity of 38.2 percent. The investigators positioned this objective instrument as a preliminary triage tool designed to help resource-constrained health examination centers prioritize patients for full overnight polysomnography rather than as a standalone diagnostic or rule-out test.

The rationale for establishing a questionnaire-free triage model stems from significant structural and operational deficits in sleep medicine across primary care settings in China. Although the country shoulders the largest absolute burden of obstructive sleep apnea worldwide, fewer than 20 percent of affected individuals receive a formal clinical diagnosis. Obstructive sleep apnea is an established independent risk factor for systemic hypertension, coronary heart disease, and all-cause mortality, making early identification clinically vital. However, access to polysomnography—the gold standard diagnostic modality—satisfies less than 5 percent of clinical demand within Chinese primary care networks. Furthermore, conventional screening instruments such as the STOP-Bang questionnaire were developed in Western populations and encounter notable linguistic, cultural, and subjective recall barriers when administered in routine Chinese health checkup environments. Because patients often struggle with subjective symptom reporting or recall bias regarding witnessed apneas and snoring intensity, researchers sought an entirely objective, low-cost screening alternative built exclusively from standardized, routinely measured physical and metabolic metrics.

The primary development cohort comprised 271 adult snoring patients who completed overnight polysomnography at the Health Management Center of the People’s Hospital of Guangxi Zhuang Autonomous Region between June 2018 and June 2024. Patients were categorized by disease severity according to their apnea-hypopnea index, with severe obstructive sleep apnea defined as having 30 or more apnea or hypopnea events per hour. Under this threshold, 143 individuals met the criteria for severe obstructive sleep apnea, while 128 presented with mild-to-moderate disease. Baseline demographic comparisons revealed marked differences between the two clinical groups. Male patients constituted 88.11 percent of the severe cohort compared to 64.84 percent of the mild-to-moderate cohort, a difference that reached statistical significance with a chi-square value of 20.725. Additionally, the severe group exhibited a slightly younger profile, with a median age of 45.00 years versus 50.50 years in the mild-to-moderate group. The authors noted that this age and sex distribution likely reflected referral selection biases within hospital checkup attendees rather than broader community-wide epidemiology.

Extensive baseline profiling demonstrated distinct anthropometric and biochemical contrasts between the mild-to-moderate and severe patient groups. Individuals with severe sleep apnea presented with significantly higher body mass index values, averaging 28.09 kilograms per square meter compared to 25.26 in the less severe group, alongside significantly larger waist circumferences averaging 93.33 centimeters compared to 86.19 centimeters. Median hip circumference was also notably elevated at 100.00 centimeters compared to 96.00 centimeters. Laboratory investigations showed that patients with severe disease had significantly higher triglyceride concentrations at 1.65 millimoles per liter versus 1.28 millimoles per liter, lower high-density lipoprotein cholesterol levels at 1.19 millimoles per liter versus 1.28 millimoles per liter, and elevated triglyceride-glucose index values at 9.47 versus 9.03. Other markers, including absolute monocyte count, serum uric acid, blood creatinine, monocyte-to-high-density lipoprotein ratio, and the systemic immune-inflammation index, were also significantly elevated in the severe group. Conversely, traditional cardiovascular parameters—such as low-density lipoprotein cholesterol, total cholesterol, glycated hemoglobin, systolic blood pressure, diastolic blood pressure, and two-hour postprandial glucose—showed no statistically significant differences between cohorts.

Model construction adhered strictly to the methodological principles outlined in the TRIPOD statement and the PROBAST risk-of-bias guidelines. Candidate variables were selected a priori based on biological plausibility, physiological rationale, and literature evidence regarding upper airway collapse and metabolic disruption. The triglyceride-glucose index was calculated using the standardized mathematical formula taking the natural logarithm of fasting triglycerides in milligrams per deciliter multiplied by fasting blood glucose in milligrams per deciliter divided by two. Although age and body mass index are widely recognized markers, body mass index exhibited heavy collinearity with waist circumference, and preliminary modeling showed that adding age or body mass index yielded only marginal discriminative improvements while adding unnecessary complexity. Neck circumference was excluded because it was not routinely measured in the institutional checkup protocol. Applying binary multivariable logistic regression with forward stepwise selection yielded a final three-variable model with 143 outcome events across three predictors, producing an events-per-variable ratio of 47.7, substantially exceeding recommended statistical minimums.

The resulting multivariable logistic regression formula was established as natural log of the odds equals negative 7.560 plus 1.306 multiplied by male sex plus 0.038 multiplied by waist circumference in centimeters plus 0.351 multiplied by the triglyceride-glucose index. In this multivariable framework, male sex emerged as a potent independent predictor of severe disease, associated with an odds ratio of 3.690 and a 95 percent confidence interval spanning 1.421 to 9.583 with a p-value of 0.007. Each unit increase in the triglyceride-glucose index was independently associated with a 42.1 percent rise in the odds of severe obstructive sleep apnea, yielding an odds ratio of 1.421 and a 95 percent confidence interval from 1.028 to 1.965 with a p-value of 0.033. In contrast, waist circumference demonstrated an odds ratio of 1.039 per centimeter with a 95 percent confidence interval spanning 0.995 to 1.085 and a p-value of 0.081. Because its 95 percent confidence interval crossed unity and its p-value exceeded the 0.05 threshold, the authors explicitly emphasized that waist circumference was not a statistically significant independent predictor, but was intentionally retained due to established biological mechanisms linking central obesity to pharyngeal airway narrowing.

Within the derivation cohort, the combined three-factor model attained an area under the receiver operating characteristic curve of 0.751 with a 95 percent confidence interval from 0.675 to 0.827. This combined performance proved statistically superior to single-variable models based solely on male sex, which achieved an area under the curve of 0.657, or the triglyceride-glucose index alone, which reached an area under the curve of 0.668, while waist circumference alone yielded an area under the curve of 0.718. At an optimal Youden index cutoff of 0.574 in the derivation cohort, the combined model demonstrated a sensitivity of 74.1 percent, a specificity of 65.2 percent, a positive predictive value of 72.4 percent, a negative predictive value of 67.2 percent, and an overall accuracy of 70.1 percent. Internal validation via 1,000 bootstrap resamples yielded an optimism-corrected area under the curve of 0.739, confirming minimal statistical overfitting. Furthermore, calibration assessment demonstrated close agreement between predicted and observed probabilities across risk deciles, confirmed by a non-significant Hosmer-Lemeshow goodness-of-fit test statistic with a p-value of 0.534.

To evaluate the model’s transportability across time, the researchers conducted a pilot temporal validation in a subsequent cohort of 82 snoring patients who underwent polysomnography during 2025 at the hospital’s Cognitive and Sleep Medicine Center. In this temporal validation set, overall discriminative performance declined to an area under the curve of 0.648 with a 95 percent confidence interval spanning 0.525 to 0.770, which remained statistically significant with a p-value of 0.024. Prioritizing sensitivity over specificity to prevent missed diagnoses in primary care, investigators evaluated a lower risk probability cutoff of 0.238. At this triage threshold, the model captured 87.2 percent of severe obstructive sleep apnea cases, yielding a negative predictive value of 68.4 percent, an overall diagnostic accuracy of 66.7 percent, a positive predictive value of 66.1 percent, and a specificity of 38.2 percent. The investigators underscored that while this low-threshold strategy successfully identifies the vast majority of high-risk patients, the resulting 38.2 percent specificity generates substantial false-positive triage referrals, making confirmatory diagnostic polysomnography indispensable.

Decision curve analysis demonstrated that applying the prediction model provided a greater net clinical benefit than either empirical treat-all or treat-none referral pathways across a broad range of threshold probabilities spanning from 0.07 to 0.93. The authors substantiated their variable selection through established physiological mechanisms. Male sex carries a 3.69-fold increased risk due to recognized anatomical differences in pharyngeal airway fat deposition and testosterone-mediated modulation of upper airway muscle tone. The triglyceride-glucose index serves as an integrated surrogate for occult insulin resistance, a metabolic derangement that correlates with escalating respiratory disturbance and sustained predictive value even among patients managed without metabolic pharmacotherapy. Regarding anthropometry, the authors cited an external polysomnography study of 339 patients where waist circumference correlated significantly with the apnea-hypopnea index with a correlation coefficient of 0.184, whereas body mass index failed to reach significance with a correlation coefficient of 0.086. However, that comparison study also revealed that the subjective Epworth Sleepiness Scale achieved a much stronger correlation of 0.649, illustrating that objective anthropometric-metabolic models often exhibit lower raw discrimination than questionnaires containing symptom scales.

The researchers acknowledged several crucial limitations that constrain immediate broad-scale implementation and require careful clinical interpretation. First, the modest temporal validation discrimination of 0.648 and low specificity of 38.2 percent preclude using the model as a standalone diagnostic tool or definitive rule-out instrument. Second, because both derivation and temporal validation cohorts originated from a single medical center in Guangxi, the findings cannot be considered externally validated across distinct geographical or administrative healthcare networks with diverse metabolic profiles. Third, omitting subjective symptoms such as witnessed apneas and excessive daytime sleepiness intentionally reduced recall bias but likely impaired discriminative power relative to hybrid screening tools. Fourth, the pronounced male predominance of 88.11 percent and younger average age in the severe group reflected referral selection bias within health checkup attendees rather than general population epidemiology. The study authors concluded that while their questionnaire-free model offers a feasible, objective triage aid to prioritize scarce polysomnography resources in primary health examination centers, multi-center prospective studies directly comparing it against established instruments like STOP-Bang are essential before widespread adoption.

Subject of Research: Clinical prediction model for severe obstructive sleep apnea using routine checkup parameters

Article Title: Development and pilot temporal validation of a questionnaire-free model for severe obstructive sleep apnea using sex, waist circumference, and TyG index

Article References: Fu, Y., Xie, Y., Wang, Z., Liao, Y., Zhou, L., He, Y., Liu, H., & Huang, J. (2026). Development and pilot temporal validation of a questionnaire-free model for severe obstructive sleep apnea using sex, waist circumference, and TyG index. Frontiers in Medicine, 13, Article 1979064. https://doi.org/10.3389/fmed.2026.1979064

Image Credits: AI Generated

DOI: 10.3389/fmed.2026.1979064

Keywords: obstructive sleep apnea, polysomnography, triglyceride-glucose index, waist circumference, primary care screening, clinical prediction model, apnea-hypopnea index, temporal validation, visceral adiposity, insulin resistance, decision curve analysis, TRIPOD guidelines, PROBAST checklist, health examination centers, multivariable logistic regression, triage screening tool, People's Hospital of Guangxi Zhuang Autonomous Region