Family history of diabetes and glycemic progression: A propensity score-based analysis using health checkup data

Figures
Abstract
Background
Family history of diabetes mellitus (FH_ DM) is a well-established risk factor for diabetes, but most studies have focused on disease incidence rather than glycemic changes over time. Understanding how FH_ DM affects the magnitude of change in blood glucose biomarkers before clinical diagnosis can inform earlier intervention strategies and optimize screening intervals.
Methods
We analyzed standardized health checkup data from Korean adults aged 20 years or older with repeated measurements over approximately 2-year (n = 25,647) and 4-year (n = 12,831) intervals. Using propensity score matching (PSM), inverse probability of treatment weighting (IPTW), and doubly robust (DR) estimation, we estimated the average treatment effect of FH_ DM on changes in fasting blood glucose (FBG) and hemoglobin A1c (HbA1c). The analytic cohort included individuals with normoglycemia, prediabetes, and diabetes. Sensitivity analyses using robustness values assessed potential unmeasured confounding.
Findings
After adjusting for 49 measured confounders, individuals with FH_ DM showed consistently greater glycemic progression than those without FH_ DM across all methods. Over 2 years, HbA1c showed a 0.02% greater increase (95% CI: 0.01–0.03) and FBG showed a 0.3 mg/dL greater increase (95% CI: 0.1–0.6) in the FH_ DM group. Over 4 years, the difference in FBG change was 1.3–1.9 mg/dL. Results remained robust when excluding individuals on diabetes medication and across sensitivity analyses.
Conclusions
Family history of diabetes is independently associated with greater glycemic progression across the normoglycemia-to-diabetes spectrum, even after rigorous adjustment for confounding. These findings support the use of FH_ DM as a practical marker for risk-stratified screening strategies. Individuals with FH_ DM may benefit from more frequent glucose monitoring and earlier preventive interventions to delay or prevent diabetes onset.
Citation: Oh S, Kim S, Yun J, Min JK, Jee H-J, Heo T-Y (2026) Family history of diabetes and glycemic progression: A propensity score-based analysis using health checkup data. PLoS One 21(6): e0352348. https://doi.org/10.1371/journal.pone.0352348
Editor: Hidetaka Hamasaki, Japanese Academy of Health and Practice, JAPAN
Received: November 3, 2025; Accepted: June 9, 2026; Published: June 24, 2026
Copyright: © 2026 Oh et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All relevant data are within the manuscript and its Supporting Information files.
Funding: This research was supported by the Bio & Medical Technology Development Program of the National Research Foundation of Korea (NRF), funded by the Korean government (MSIT) (No. RS-2024-00440787), and by an additional NRF grant funded by the Korean government (MSIT) (No. RS-2024-00354881). This research was also supported by the Regional Innovation System & Education (RISE) program through the Chungbuk Regional Innovation System & Education Center, funded by the Ministry of Education (MOE) and the Chungcheongbuk-do, Republic of Korea (2026-RISE-11-014-03). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Diabetes mellitus has become one of the most serious public health challenges in South Korea. The number of individuals diagnosed with diabetes has recently exceeded six million. When those with prediabetes are included, the number of people who either have diabetes or are at risk is estimated to be over twenty million, representing nearly forty percent of the adult population [1]. This heavy burden emphasizes the urgent need for effective strategies that focus on prevention, early detection, and timely intervention. Diabetes can be managed through proper treatment and healthy lifestyle modifications, and the World Health Organization (WHO) has emphasized that early screening for non-communicable diseases is highly cost-effective.
The prevalence of diabetes is rising steadily, which means that the number of people with a family history of diabetes is also expected to increase. Family history is widely recognized as an important risk factor for diabetes. It reflects the combined influence of inherited genetic factors, shared lifestyle and environmental conditions such as diet, exercise, and socioeconomic status, and even possible epigenetic mechanisms [2–6]. Because of these features, family history has long been used as a simple and practical way to identify individuals at higher risk of developing diabetes. At the same time, family history can be subject to reporting errors, and with the growing availability of precision medicine and genetic testing, its role as a proxy measure is becoming more complex.
The diagnosis of diabetes and prediabetes is usually based on fasting blood glucose (FBG) and hemoglobin A1c (HbA1c). These two biomarkers represent different aspects of glucose metabolism. FBG provides information on short-term glucose control and hepatic glucose production, while HbA1c reflects the average blood glucose concentration over the preceding two to three months [7,8]. Examining how family history affects these continuous markers can provide a more detailed understanding of the pathways through which diabetes develops. This type of analysis can also capture early metabolic abnormalities that occur before a clinical diagnosis is made.
Most previous studies have investigated family history in relation to diabetes as a binary outcome [9,10]. Many of these studies have relied on self-reported family history and used cross-sectional regression models or machine learning approaches. Such methods can identify associations, but they may fail to address confounding properly and cannot describe the relationship between family history and glucose regulation in detail. Recent studies have examined family history effects on diabetes complications in diagnosed patients [10], but few have investigated whether family history of diabetes (FH_ DM) affects glycemic progression before clinical diagnosis. Understanding the magnitude of glycemic change in individuals with FH_ DM across the normoglycemia-to-diabetes spectrum is critical for informing screening interval decisions and early intervention strategies.
The present study has several strengths that differentiate it from earlier work. First, we use continuous outcomes, namely FBG and HbA1c, rather than relying only on a binary diabetes diagnosis. This enables us to detect differences in glycemic progression that may precede the onset of diabetes. Second, we analyze high-quality laboratory data from standardized national health checkups in South Korea. These checkups provide objective and comprehensive information that includes biochemical and clinical measurements, which reduces the risk of bias from self-reported data [11]. Third, we apply propensity score methods to estimate the average treatment effect of family history, which allows us to approximate the balance achieved in randomized trials and to provide more robust causal estimates. The use of multiple propensity score approaches (including matching, weighting, and doubly robust estimation) has been successfully applied in diabetes research [10,12] and strengthens confidence in our findings.
The aim of this study is to evaluate the effect of family history of diabetes on changes in FBG and HbA1c levels. By focusing on continuous biomarkers and by applying rigorous causal inference methods, this research seeks to clarify whether individuals with a family history of diabetes display different patterns of glycemic progression compared with those without such a history. The findings are expected to contribute evidence that will inform public health policies and support strategies for the early prevention and management of diabetes.
Materials and methods
Study design and data source
Participants were adults aged 20 years or older who underwent standardized health checkups. Health checkup data were organized into repeated measurement periods of approximately two years each. We defined glycemic progression as the change in fasting blood glucose (FBG) and hemoglobin A1c (HbA1c) between periods. The difference between the first and second checkups was defined as a one-period (approximately 2-year) change, and the difference between the first and third checkups as a two-period (approximately 4-year) change (Fig 1).
Each interval is approximately 2 years. The arrow at the top indicates the 4-year difference between the 1st and 3rd visits (Period 2), while the arrows at the bottom denote the 2-year differences between consecutive visits (Period 1).
The analytic cohort included individuals with normoglycemia, prediabetes, and diabetes. In Period 1 (2-year interval), 25,647 individuals were included: 3,149 with family history of diabetes (FH_ DM) and 22,498 without FH_ DM. In Period 2 (4-year interval), 12,831 individuals were included: 1,661 with FH_ DM and 11,170 without FH_ DM. To assess robustness, we performed a sensitivity analysis excluding individuals prescribed diabetes medication, which reduced the samples to 25,560 (Period 1) and 12,747 (Period 2) individuals.
Exposure and outcome definitions
Family history of diabetes mellitus (FH_ DM) was defined as a binary indicator of whether participants self-reported any family history of diabetes in their health checkup records. This served as the exposure variable, with individuals reporting FH_ DM comprising the treated group and those without FH_ DM comprising the untreated group.
The primary outcomes were changes in two blood glucose biomarkers: fasting blood glucose (FBG, measured in mg/dL) and hemoglobin A1c (HbA1c, measured in %). These continuous outcomes represent glycemic progression over time and provide complementary information about glucose metabolism. FBG reflects short-term glucose control and hepatic glucose production, while HbA1c represents average blood glucose concentration over the preceding 2–3 months [7,8]. Because the outcomes were defined as changes from baseline, inter-individual differences in baseline glycemic status are inherently incorporated into the outcome, supporting the validity of the pooled analysis across glycemic strata.
Covariates
The dataset included 49 baseline covariates measured at the beginning of each analysis interval (Table 1). These comprised: demographic factors (age, sex); anthropometric measurements (body mass index, waist circumference, body fat percentage); vital signs (blood pressure, pulse rate); pulmonary function (forced vital capacity, FEV1/FVC ratio); complete blood count parameters; metabolic markers (lipid profile, liver enzymes, kidney function tests, electrolytes); inflammatory markers (ESR, CRP); thyroid function tests; urine analysis; Helicobacter pylori status; and medication use for hypertension, diabetes, and hyperlipidemia. Variables irrelevant to glucose metabolism (blood type, urine specific gravity) were excluded from the analysis.
Statistical analysis
Analytical framework.
We estimated the average treatment effect (ATE) of FH_ DM on glycemic progression using propensity score methods within a potential outcomes framework [13–16]. Given that family history is a non-manipulable characteristic, these estimates are interpreted as adjusted associations between individuals with and without FH_ DM rather than as effects of a literal intervention. This approach addresses confounding by balancing measured covariates between individuals with and without FH_ DM, thereby approximating the balance achieved in randomized trials.
The ATE represents the expected difference in outcomes if the entire population had FH_ DM versus if they did not. In randomized trials, ATE can be estimated as the simple difference in mean outcomes between groups. However, in observational studies, propensity score methods are needed to adjust for confounding [16,17].
This approach requires several key assumptions [18,19]: (1) no unmeasured confounding after adjusting for observed covariates (conditional exchangeability); (2) both treated and untreated individuals exist across all covariate combinations (positivity); and (3) well-defined treatment with no interference between individuals (Stable Unit Treatment Value Assumption, SUTVA). While these assumptions cannot be definitively verified, we assessed their plausibility through balance diagnostics and sensitivity analyses.
Propensity score estimation and methods.
Propensity scores (the conditional probability of having FH_ DM given baseline covariates) were estimated using logistic regression with all 49 covariates measured at the start of each period [17]. We then applied three complementary propensity score methods:
Propensity score matching (PSM): We performed 1:1 nearest-neighbor matching without replacement using a caliper width of 0.2 standard deviations of the logit propensity score [20,21]. This caliper approach has been shown to effectively reduce bias while maintaining adequate sample size. Matching created comparable treated and untreated groups, after which we estimated the ATE as the mean difference in outcomes between matched pairs.
Inverse probability of treatment weighting (IPTW): Each individual was weighted by the inverse of their probability of receiving their actual treatment status [22,23]. Specifically, treated individuals received weight and untreated individuals received weight , where X represents the covariates. This weighting creates a pseudo-population in which measured confounders are balanced across groups. We then estimated the ATE using weighted linear regression [24].
Doubly robust (DR) estimation: This method combines propensity score weighting with outcome regression modeling [25,26]. The DR estimator provides valid results if either the propensity score model or the outcome model is correctly specified, thereby offering additional protection against model misspecification.
Balance assessment and sensitivity analysis.
We assessed covariate balance using standardized mean differences, defined as the mean difference between groups divided by the pooled standard deviation [27,28]. Standardized differences below 0.1 (10%) were considered to indicate adequate balance. Balance diagnostics are presented in Fig 2 and Fig 3 for PSM and IPTW, respectively.
Vertical dashed lines indicate a standardised mean difference of 0.1.
Vertical dashed lines indicate a standardised mean difference of 0.1.
To evaluate robustness to potential unmeasured confounding, we calculated robustness values (RV) [29]. The RV quantifies the minimum proportion of residual variance that an unmeasured confounder would need to explain in both the exposure and outcome simultaneously to reduce the estimated association to zero. For example, an RV of 0.03 indicates that an unmeasured confounder would need to explain at least 3% of the residual variance in both FH_ DM and glycemic progression simultaneously to nullify our findings. Values below 0.05 suggest potential sensitivity to unmeasured confounding, and results should therefore be interpreted with appropriate caution.
All analyses were conducted using R version 4.3.1. Continuous variables were summarized as mean (SD) for normally distributed data and median (Q1, Q3) for skewed data. Categorical variables were summarized as number (%). Statistical significance was assessed using 95% confidence intervals.
Results
Study population and baseline characteristics
A total of 25,647 participants were included in Period 1 (2-year interval) and 12,831 in Period 2 (4-year interval). Among them, 3,149 (12.3%) and 1,661 (12.9%) participants, respectively, reported a family history of diabetes (FH_ DM).
Baseline characteristics are presented in Table 1. Before adjustment, participants with FH_ DM were slightly older and had higher body mass index and waist circumference compared with those without FH_ DM. After applying propensity score methods, all 49 covariates achieved standardized mean differences below 0.1 in both PSM and IPTW, indicating adequate balance between groups (Figs 2 and 3).
Average treatment effects of family history on glycemic progression
Table 2 presents the estimated average treatment effects (ATE) of FH_ DM on changes in HbA1c and FBG. Results were generally consistent across methods, with IPTW and DR showing similar effect sizes. PSM yielded larger estimates in the 2-period analysis, particularly for FBG, and was therefore interpreted as supportive evidence rather than the primary reference for effect magnitude.
HbA1c changes: Over 2 years, individuals with FH_ DM showed a 0.02% greater increase in HbA1c compared to those without FH_ DM, with consistent results across all three methods (PSM: 0.020%, 95% CI: 0.014–0.026; IPTW: 0.021%, 95% CI: 0.013–0.030; DR: 0.022%, 95% CI: 0.009–0.035). Over 4 years, IPTW and DR continued to show similar effects (0.017–0.020%), while PSM indicated a larger effect (0.034%, 95% CI: 0.023–0.045).
FBG changes: The adjusted differences in FBG were more pronounced and increased substantially over longer follow-up. Over 2 years, the difference in FBG change between groups was approximately 0.3 mg/dL (PSM: 0.289 mg/dL, 95% CI: 0.090–0.487; IPTW: 0.293 mg/dL, 95% CI: 0.013–0.573; DR: 0.299 mg/dL, 95% CI: −0.125–0.757). Over 4 years, this difference increased substantially to 1.3–1.9 mg/dL across methods (PSM: 1.912 mg/dL, 95% CI: 1.596–2.227; IPTW: 1.317 mg/dL, 95% CI: 0.889–1.745; DR: 1.361 mg/dL, 95% CI: 0.712–2.079).
Across both outcomes and time periods, individuals with FH_ DM consistently showed greater glycemic progression than those without FH_ DM. The consistency of findings across IPTW and DR methods, which employ different analytical strategies, strengthens confidence in these results. Of note, the between-group difference in FBG change increased approximately four- to six-fold from the 2-year to the 4-year period, suggesting an acceleration of glycemic divergence over time. In contrast, HbA1c differences remained comparatively stable across periods, reflecting a more gradual and sustained pattern of progression.
Sensitivity analyses
Excluding individuals on diabetes medication yielded highly similar results (Table 3), confirming that our findings were not driven by pharmacological treatment effects. In this medication-excluded analysis, the pattern of results remained consistent: HbA1c showed differences of 0.016–0.025% and FBG showed differences of 0.343–1.906 mg/dL depending on the period and method used.
Additional stratified analyses by baseline glycemic status were performed to assess whether the pooled analysis was supported across glycemic categories (S1 Table). The positive association between FH_ DM and glycemic progression was generally observed among participants with normoglycemia and prediabetes, particularly for FBG over 4 years. Estimates in the diabetes subgroup were imprecise because of the small sample size, especially in the 4-year analysis.
Robustness value analysis (Table 4) indicated that unmeasured confounders would need to explain 1.1–4.4% of the residual variance in both FH_ DM and the outcomes to nullify our findings. For example, the robustness value for HbA1c in the 2-year analysis was 0.027, meaning an unmeasured confounder would need to explain at least 2.7% of the residual variance in both treatment and outcome to reduce the estimated association to zero. While these relatively low values suggest potential sensitivity to unmeasured confounding, the comprehensive adjustment for 49 measured covariates and consistency of results across multiple methods support the validity of our estimates.
Discussion
This study provides evidence that family history of diabetes mellitus (FH_ DM) is associated with greater glycemic progression, even after rigorous adjustment for measured confounders. Using multiple propensity score methods applied to high-quality health checkup data, we found that individuals with FH_ DM showed consistently larger increases in both HbA1c and fasting blood glucose compared to those without FH_ DM.
The observed differences in glycemic progression, while modest in absolute terms at the individual level, carry meaningful implications at the population level. While a 0.02% greater increase in HbA1c over 2 years would not in isolation shift a single patient’s clinical management, across a large population of individuals with family history of diabetes, this difference represents a substantial shift in the overall distribution of glycemic risk, with implications for screening policy and public health resource allocation. An HbA1c difference of approximately 0.02% over 2 years in individuals with FH_ DM translates to roughly 0.1% over 10 years. This projection, however, assumes a constant rate of progression, which may not hold over longer periods due to regression to the mean, lifestyle modifications, or pharmacological interventions. Longer-term studies are therefore needed to confirm whether the observed differences persist and accumulate over time. Given that the threshold for diabetes diagnosis is HbA1c > 6.5% and for prediabetes is 5.7–6.4% [7], this accelerated progression could shift individuals across diagnostic categories several years earlier than those without FH_ DM. The pattern for FBG was even more pronounced, with differences increasing from approximately 0.3 mg/dL over 2 years to 1.3–1.9 mg/dL over 4 years, suggesting that individuals with FH_ DM may reach prediabetic or diabetic FBG levels substantially earlier [6,7].
Our findings align with and extend previous epidemiological evidence. The Framingham Offspring Study demonstrated that family history nearly doubled the risk of type 2 diabetes [30,31], while other studies have documented associations between FH_ DM and various metabolic abnormalities [2–6]. However, most prior research focused on diabetes incidence as a binary outcome. By examining continuous biomarkers and their magnitude of change over time, our study captures earlier metabolic deterioration that precedes clinical diagnosis. Our findings also complement recent work by Plattner et al. [10], who examined post-diagnosis complications, whereas we investigated pre-diagnostic glycemic progression, together suggesting that the association of family history spans the entire diabetes natural history.
These findings have direct relevance for diabetes screening policies. Current guidelines in many countries, including South Korea, recommend routine diabetes screening every 3 years for average-risk adults [1]. However, our results suggest that individuals with FH_ DM may benefit from more frequent monitoring, such as annual or biennial glucose testing. Such a risk-stratified approach enables earlier detection when interventions are most effective, using a readily available risk marker that requires no additional testing, and may prove cost-effective by focusing intensive monitoring on high-risk individuals [3]. In South Korea, where standardized health checkups are widely available, systematically utilizing family history information could enhance diabetes prevention efforts at the population level.
The consistency of results across IPTW and doubly robust approaches strengthens confidence in our findings, with the doubly robust method providing additional protection against model misspecification [25,26]. In the 4-year analysis, PSM produced larger estimates than IPTW and DR, likely because matching restricts the analysis to individuals with sufficient covariate overlap and thereby changes the analytic population. Thus, the PSM estimates may reflect effects in the matched subset rather than the overall study population. We therefore interpreted the PSM results as supportive sensitivity evidence and used the IPTW and DR estimates as the primary reference for the magnitude of association. Sensitivity analyses using robustness values indicated that unmeasured confounders would need to explain 1.1–4.4% of residual variance in both FH_ DM and the outcomes to nullify our findings, suggesting caution in interpretation. Nevertheless, the comprehensive adjustment for 49 measured covariates and consistency across methods support the validity of our estimates.
The stratified analysis by baseline glycemic status further supported the pooled analysis. The association between FH_ DM and glycemic progression was most consistent among individuals with normoglycemia or prediabetes, whereas estimates among those with diabetes were unstable because of limited sample size, particularly in the 4-year analysis, and may have been influenced by more intensive monitoring, lifestyle modification, or pharmacological management after diabetes recognition.
Several limitations warrant consideration. First, family history was self-reported and treated as a binary variable. The health checkup questionnaire did not distinguish between first-degree and more distant relatives, and underreporting would likely attenuate our estimates toward the null. Second, residual confounding from unmeasured factors such as diet, physical activity, and genetic polymorphisms remains possible, all of which are known to influence glucose metabolism [32]. Third, our study population was drawn from a single hospital’s health checkup program in Korea, which may limit generalizability to other populations. Fourth, the observational design cannot definitively establish causality, as family history reflects both genetic predisposition and shared environmental factors [2–5]. Finally, follow-up periods of 2 and 4 years are relatively short, and longer-term studies are needed to assess whether the observed differences ultimately translate to earlier diabetes onset.
Future research should incorporate more detailed family history information, lifestyle and genetic data, and multi-center designs across diverse populations to address these limitations and confirm the long-term implications of our findings.
Conclusions
Using rigorous propensity score methods applied to longitudinal health checkup data, we found that family history of diabetes is associated with greater glycemic progression. Individuals with FH_ DM showed consistently larger increases in HbA1c and fasting blood glucose across multiple analytical approaches.
These findings support the use of FH_ DM as a practical risk marker for identifying individuals who may benefit from intensified monitoring and earlier preventive interventions. Given that family history information is readily obtained during routine health checkups without additional cost, integrating FH_ DM into risk-stratified screening strategies represents an actionable approach to diabetes prevention.
As diabetes prevalence continues rising globally, optimizing screening and prevention strategies for high-risk populations is increasingly important. Our results contribute evidence that can inform clinical guidelines and public health policies, ultimately supporting efforts to reduce the burden of diabetes through earlier detection and intervention.
Supporting information
S1 File. Anonymized dataset for Period 1 (2-year interval) and Period 2 (4-year interval).
https://doi.org/10.1371/journal.pone.0352348.s001
(ZIP)
S1 Table. Stratified sensitivity analysis of the adjusted association of family history of diabetes with glycemic progression by baseline glycemic status.
Values represent the difference in change in HbA1c (%) or FBG (mg/dL) between individuals with family history of diabetes and those without, with 95% confidence intervals in parentheses. Glycemic status was classified according to American Diabetes Association criteria. n, sample size (FH_DM / non-FH_DM); PSM, propensity score matching; IPTW, inverse probability of treatment weighting; DR, doubly robust estimation.
https://doi.org/10.1371/journal.pone.0352348.s002
(PDF)
Acknowledgments
This manuscript is based on the master’s dissertation of Sangtaek Oh, submitted to Chungbuk National University in 2025.
References
- 1. Kwon H-S. Prevalence and treatment status of diabetes mellitus in Korea. J Korean Med Assoc. 2023;66(7):404–7.
- 2. Groop L, Lyssenko V. Genes and type 2 diabetes mellitus. Curr Diab Rep. 2008;8(3):192–7. pmid:18625115
- 3. Yoon PW, Scheuner MT, Khoury MJ. Research priorities for evaluating family history in the prevention of common chronic diseases. Am J Prev Med. 2003;24(2):128–35. pmid:12568818
- 4. Annis AM, Caulder MS, Cook ML, Duquette D. Family history, diabetes, and other demographic and risk factors among participants of the National Health and Nutrition Examination Survey 1999-2002. Prev Chronic Dis. 2002;2.
- 5. Sterns JD, Smith CB, Steele JR, Stevenson KL, Gallicano GI. Epigenetics and type II diabetes mellitus: underlying mechanisms of prenatal predisposition. Front Cell Dev Biol. 2014;2:15. pmid:25364722
- 6. Hu S, Ji W, Zhang Y, Zhu W, Sun H, Sun Y. Risk factors for progression to type 2 diabetes in prediabetes: a systematic review and meta-analysis. BMC Public Health. 2025;25(1):1220. pmid:40165126
- 7. Tabák AG, Herder C, Rathmann W, Brunner EJ, Kivimäki M. Prediabetes: a high-risk state for diabetes development. Lancet. 2012;379(9833):2279–90. pmid:22683128
- 8. Kim E-H, Bae S-G, Kim K-Y, Na Y. Impaired Fasting Glucose Rate by Diabetes Family History. J Health Info Stat. 2017;42(1):63–9.
- 9. Zhang J, Yang Z, Xiao J, Xing X, Lu J, Weng J, et al. Association between family history risk categories and prevalence of diabetes in Chinese population. PLoS One. 2015;10(2):e0117044. pmid:25664814
- 10. Plattner C, Pfeifer B, Sourij H, Vill D, Wiedl M, Middeldorf K, et al. Effects of family history and sex on diabetes-related outcome in type 2 diabetes - Analysis from the tyrolean diabetes registry. PLoS One. 2025;20(6):e0324696. pmid:40531860
- 11. Yuk H, Gim J, Min JK, Yun J, Heo TY. Artificial intelligence-based prediction of diabetes and prediabetes using health checkup data in Korea. Appl Artif Intell. 2022;36(1):2145644.
- 12. Tao B, Pietropaolo M, Atkinson M, Schatz D, Taylor D. Estimating the cost of type 1 diabetes in the U.S.: a propensity score matching method. PLoS One. 2010;5(7):e11501. pmid:20634976
- 13. Holland PW. Statistics and causal inference. J Am Stat Assoc. 1986;81(396):945–60.
- 14. Ramsey DSL, Forsyth DavidM, Wright E, McKay M, Westbrooke I. Using propensity scores for causal inference in ecology: Options, considerations, and a case study. Methods Ecol Evol. 2019;10(3):320–31.
- 15.
Morgan S. Counterfactuals and Causal Inference. Cambridge: Cambridge University Press; 2015.
- 16. Austin PC. An Introduction to Propensity Score Methods for Reducing the Effects of Confounding in Observational Studies. Multivariate Behav Res. 2011;46(3):399–424. pmid:21818162
- 17. ROSENBAUM PR, RUBIN DB. The central role of the propensity score in observational studies for causal effects. Biometrika. 1983;70(1):41–55.
- 18. Williamson E, Morley R, Lucas A, Carpenter J. Propensity scores: from naive enthusiasm to intuitive understanding. Stat Methods Med Res. 2012;21(3):273–93. pmid:21262780
- 19. Shiba K, Kawahara T. Using Propensity Scores for Causal Inference: Pitfalls and Tips. J Epidemiol. 2021;31(8):457–63. pmid:34121051
- 20. Rosenbaum PR, Rubin DB. Constructing a control group using multivariate matched sampling methods that incorporate the propensity score. Am Stat. 1985;39(1):33–8.
- 21. Austin PC. A comparison of 12 algorithms for matching on the propensity score. Stat Med. 2014;33(6):1057–69. pmid:24123228
- 22. Robins J. A new approach to causal inference in mortality studies with a sustained exposure period—application to control of the healthy worker survivor effect. Math Model. 1986;7(9–12):1393–512.
- 23. Chesnaye NC, Stel VS, Tripepi G, Dekker FW, Fu EL, Zoccali C, et al. An introduction to inverse probability of treatment weighting in observational research. Clin Kidney J. 2021;15(1):14–20. pmid:35035932
- 24.
Hernán MA, Robins JM. Causal Inference: What If. Boca Raton (FL): CRC Press; 2020.
- 25.
Facure M. Causal Inference for the Brave and True [Internet]. 2023 [cited 2025 January 15]. Available from: https://matheusfacure.github.io/python-causality-handbook/landing-page.html
- 26. Bang H, Robins JM. Doubly robust estimation in missing data and causal inference models. Biometrics. 2005;61(4):962–73. pmid:16401269
- 27. Austin PC. Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples. Stat Med. 2009;28(25):3083–107. pmid:19757444
- 28. Williamson EJ, Forbes A. Introduction to propensity scores. Respirology. 2014;19(5):625–35. pmid:24889820
- 29. Cinelli C, Hazlett C. Making Sense of Sensitivity: Extending Omitted Variable Bias. J R Stat Soc Series B Stat Methodol. 2020;82(1):39–67.
- 30. Wilson PWF, Meigs JB, Sullivan L, Fox CS, Nathan DM, D’Agostino RB Sr. Prediction of incident diabetes mellitus in middle-aged adults: the Framingham Offspring Study. Arch Intern Med. 2007;167(10):1068–74. pmid:17533210
- 31. Franks PW. Diabetes family history: a metabolic storm you should not sit out. Diabetes. 2010;59(11):2732–4. pmid:20980473
- 32. Codazzi V, Frontino G, Galimberti L, Giustina A, Petrelli A. Mechanisms and risk factors of metabolic syndrome in children and adolescents. Endocrine. 2024;84(1):16–28. pmid:38133765