Skip to main content
NCBI home page
Wiley Open Access Collection logoLink to Wiley Open Access Collection
. 2026 Jan 27;50(1):e70028. doi: 10.1002/gepi.70028

Integrative Harmonization of Phenotypic and Genomic Data Improves Bone Mineral Density Prediction in Multi‐Study Osteoporosis Research

Anqi Liu 1, Jianing Liu 1, Lang Wu 2,3, Qing Wu 1,✉
PMCID: PMC12836449  PMID: 41589508

ABSTRACT

Harmonizing osteoporosis‐related data across multiple data sets is essential for improving the accuracy and generalizability of bone mineral density (BMD) assessments. This study developed a harmonization framework to standardize phenotypic and genomic variables across three major US osteoporosis data sets: GDBF, GWAS, and NHANES. We standardized key phenotypic variables (BMD, body mass index (BMI), age, sex, and race/ethnicity) using cohort‐specific data dictionaries and applied multiple imputations by chained equations (MICEs) to manage missing data. Genomic data were harmonized using principal component analysis (PCA)‐based batch effect corrections. Residual regression methods were applied to standardize BMD values. The effectiveness of harmonization on BMD prediction was evaluated using generalized estimating equations (GEEs) and mixed‐effects models. Post‐harmonization, inter‐study variability in BMI was significantly reduced (Ω 2 = 0.0028), and BMD associations with covariates remained consistent across data sets. Harmonized models showed improved predictive performance, with explained variance in BMD increasing (R 2 = 0.14%). PCA confirmed the effective alignment of genetic data, reducing batch effects and improving cross‐study compatibility. This study demonstrates the feasibility and effectiveness of harmonizing phenotypic and genomic data for osteoporosis research. The harmonization framework enhances BMD prediction accuracy, supports more inclusive osteoporosis risk assessment, and improves the integration of multi‐cohort data sets for future research. These findings highlight the potential of data harmonization in advancing precision medicine for osteoporosis prevention and management.

Keywords: bone mineral density (BMD), data harmonization, fracture risk prediction, genomic integration, osteoporosis, phenotypic standardization

1. Introduction

Osteoporosis, a chronic condition characterized by reduced bone mineral density (BMD) and increased fracture risk, affects approximately 18.3% of the global population, with 1 in 3 women and 1 in 5 men over 50 predicted to experience an osteoporotic fracture (Sozen et al. 2017; Kanis et al. 2019). Despite its prevalence and significant burden on public health, osteoporosis often remains underdiagnosed and undertreated, contributing to avoidable morbidity and mortality globally (Singer et al. 2023; Cummings and Melton 2002).

Genetic studies, particularly genome‐wide association studies (GWAS), have identified numerous single‐nucleotide polymorphisms (SNPs) associated with BMD and fracture risk, shedding light on the genetic basis of osteoporosis (Yuan et al. 2019; Estrada et al. 2012; Zheng et al. 2015). However, existing research is limited by challenges in harmonizing genomic and phenotypic data across diverse studies (Brookes and Robinson 2015; Fortier et al. 2019). Variability in data collection methods, demographic representation, and the lack of standardized protocols for defining key variables such as BMD and body mass index (BMI) impede the ability to perform robust, large‐scale analyses (Luningham et al. 2019). For example, inconsistencies in SNP imputation, differences in measurement techniques, and underrepresentation of minority populations have limited the generalizability of findings to non‐European ancestry groups (Albrechtsen et al. 2010; Hinrichs et al. 2014; Landry et al. 2018; Popejoy and Fullerton 2016).

Harmonizing phenotypic and genomic data is critical to overcoming these barriers and advancing osteoporosis research. By aligning variable definitions, standardizing measurement units, and addressing demographic imbalances, harmonized data sets enable more comprehensive analyses of gene‐environment interactions, improve statistical power, and foster equitable research outcomes (Pan et al. 2023; Wong et al. 2003; Stilp et al. 2021; Sung et al. 2019). However, Previous harmonization efforts often lacked comprehensive approaches to simultaneously standardizing both phenotypic and genomic variables, limiting their generalizability and practical applicability. Significant obstacles remain, including biases introduced by missing data, measurement instrument incompatibility, and varying inclusion criteria across data sets (Schmidt et al. 2021; Little and Rubin 2019).

This study aimed to address these challenges by implementing a harmonization framework across multiple major US osteoporosis data sets. The framework focuses on:

  • 1.

    Standardizing phenotypic variables, such as BMD, BMI, and demographic factors, across data sets.

  • 2.

    Applying robust statistical techniques, including mixed‐effects models and generalized estimating equations (GEEs), to account for inter‐study variability.

  • 3.

    Harmonizing genomic data through stringent quality control and imputation protocols to enable integration across data sets.

By providing a harmonized data set that integrates both phenotypic and genomic data, we aim to establish a foundation for harmonized large‐scale investigations into the phenotypic and genetic determinants of osteoporosis. The insights derived from this harmonization framework have the potential to enhance the understanding of BMD variation across diverse populations, advancing the field toward more inclusive and effective strategies for osteoporosis prevention and management.

2. Methods

2.1. Study Design and Data Sources

This study harmonized data from three major US osteoporosis studies and data sets: the Genetic Determinants of Bone Fragility (GDBF) study, the GWAS for Osteoporosis, and the National Health and Nutrition Examination Survey (NHANES). These data sets were selected because they include key bone health measures, such as BMD, and possess comprehensive phenotypic variables and genomic data for osteoporosis research, enabling a multi‐data set evaluation of phenotype harmonization.

The GDBF study is a cross‐sectional investigation assessing genetic factors influencing peak BMD in premenopausal women (National Center for Biotechnology Information n.d–b). The GWAS study is a longitudinal cohort designed to explore genome‐wide associations with osteoporosis, emphasizing genetic markers relevant to bone health (National Center for Biotechnology Information n.d–a). NHANES is a nationally representative, cross‐sectional survey conducted by the National Center for Health Statistics, providing data on BMD and other health parameters (Akinbam et al. 2022). This study specifically utilized data from the 2009–2010, 2013–2014, and 2017–2020 cycles (Liang 2023). The overall harmonization process, including both phenotypic and genomic data, is summarized in Figure 1, which illustrates the multistep approach used in this study, including multi‐study phenotype harmonization and BMD‐specific harmonization.

Figure 1.

Figure 1

Open in a new tab

Harmonization pipeline for phenotypic and genomic data. Phenotypic harmonization involved data cleaning (handling missing values and outlier detection), followed by standardization and statistical analyses (Kruskal–Wallis test and residual regression) to ensure consistency across cohorts. Genomic harmonization of GWAS and GDBF data sets included rigorous quality control, imputation, and batch‐effect correction, facilitating accurate integration and analysis.

Data were obtained from both controlled‐access and publicly available repositories. The GDBF and GWAS data sets were accessed through the Database of Genotypes and Phenotypes (dbGaP) (dbGaP study accession numbers: phs000390.v1.p1 and phs000138.v2.p1). The NHANES was obtained from the National Center for Health Statistics (NCHS). Ethical approval was obtained from the relevant institutional review board (IRB #2025E0335).

2.2. Phenotypic Harmonization

In this study, BMD specifically refers to total hip BMD measured by dual‐energy X‐ray absorptiometry (DXA) and expressed in grams per square centimeter (g/cm2). BMD values in the GDBF cohort were pre‐adjusted in dbGaP using residuals derived from linear regression models adjusting for baseline age and weight. To align methodologically with the GDBF adjustments, we applied the same residual regression approach separately within the GWAS cohort, replacing original BMD values with residuals computed from models adjusting for baseline age and weight. NHANES lacked hip BMD measurements and were therefore included solely for demographic comparisons.

Other key phenotypic variables—including BMI, age, sex, and race/ethnicity—were standardized using a structured harmonization framework. Variable definitions were consistently aligned across data sets through cohort‐specific data dictionaries. BMI values were uniformly recalculated from height and weight measurements. Additionally, racial and ethnic categories were consolidated into four standardized groups: Black, Hispanic, White, and Other.

2.2.1. Handling of Missing Data and Imputation

Multiple imputations by chained equations (MICEs) were applied separately within each cohort, assuming missing‐at‐random (MAR) (White et al. 2011; Azur et al. 2011). The imputation model included age, sex, BMI, race, and BMD to ensure unbiased estimates. Convergence diagnostics were performed, and post‐imputation validation confirmed that the imputed values were closely aligned with observed data distributions.

2.3. Genomic Harmonization

Genomic data from GWAS and GDBF studies underwent rigorous quality control and harmonization.

2.3.1. Genotyping and Imputation

Genotyping was conducted using Illumina and Affymetrix arrays (the Haplotype Reference Consortium 2016), followed by genotype imputation via the Haplotype Reference Consortium (HRC) panel using the Michigan Imputation Server (Durbin 2014). Quality control measures included the removal of SNPs with a minor allele frequency (MAF) < 0.01, call rate < 99%, or Hardy–Weinberg equilibrium (HWE) p value < 1 × 10−5. Imputation accuracy for genetic data was evaluated using INFO scores and posterior genotype dosage R 2 values, both provided by the Michigan Imputation Server.

2.3.2. Population Structure and Batch Effect Correction

Principal component analysis (PCA) was used to evaluate population structure and detect batch effects (Abdi and Williams 2010; Price et al. 2006). Scatter plots of PC1 versus PC2 confirmed genetic similarity between the GWAS and GDBF data sets. Study‐specific genomic batch effects were assessed and corrected, ensuring consistency across data sets. Quality control, PCA, and batch‐effect corrections were performed using PLINK 2.0 and R statistical software (version 4.3.2), ensuring transparency and reproducibility.

2.4. Statistical Analysis

A combination of statistical approaches was applied to evaluate BMD associations and harmonization effectiveness. Prior to statistical analysis, model assumptions—including linearity, independence, and homoscedasticity—were examined through residual diagnostic plots to assess the randomness of residuals and the constancy of variance. To assess model robustness, parallel analyses using both GEE and mixed‐effects models were conducted, and estimates were compared to evaluate consistency across modeling strategies.

2.4.1. Harmonization Effectiveness Assessment

Harmonization effectiveness was assessed by comparing the pre‐ and post‐harmonization distributions of key phenotypic variables using analysis of variance (ANOVA) and the Kruskal–Wallis test. Residual regression analyses were performed to reduce inter‐study variability, adjusting for age, sex, and race. Effect sizes (Omega squared, Ω 2) were calculated to quantify the proportion of variance explained by study differences after harmonization (Lakens 2013).

2.4.2. BMD Association Analysis

BMD was modeled as a function of age, sex, BMI, and race using GEE (Liang and Zeger 1986) for harmonized data, which accounted for the within‐study correlation. Mixed‐effects models were implemented (Laird and Ware 1982), incorporating the study cohort as a random effect to capture inter‐data variability. Additionally, separate linear regression models were applied to pre‐harmonized GWAS and GDBF data sets to compare association estimates before and after harmonization.

2.4.3. Model Fit and Validation

Model performance was evaluated by comparing estimates across different statistical approaches. Adjusted R 2 (for linear regression), Pseudo R 2 (for mixed‐effects models), and Quasi R 2 (for GEEs) were used to assess predictive accuracy, providing insights into model selection and comparative effectiveness.

3. Results

3.1. Data Set Characteristics

The final harmonized data set included 5663 participants from three major data sets: GDBF (n = 1493), GWAS (n = 1617), and NHANES (n = 2553). Table 1 presents the baseline characteristics before and after harmonization. Before harmonization, significant differences were observed across the data sets in age, BMI, sex, and racial composition (ANOVA, p < 0.01). The mean BMI varied across data sets, with GDBF participants having a lower mean BMI (26.2 kg/m2) compared with NHANES (28.2 kg/m2) and GWAS (26.8 kg/m2). Similarly, the racial composition was notably different, with NHANES containing a more diverse population (39.5% White, 29.7% Hispanic, 20.6% Black, and 10.1% Other), whereas GWAS and GDBF were predominantly White (100% and 99.6%, respectively). The GDBF data consisted of younger participants (mean age: 32.7 years), whereas GWAS and NHANES participants had higher mean ages of 48.1 and 48.8 years, respectively.

Table 1.

Participant characteristics of the NHANES, GWAS, GDBF data sets, and post‐harmonization data set.

Characteristic GDBF (N = 1493) GWAS (N = 1617) NHANES (N = 2553) Post‐harmonization (N = 5663) p **
Age (years) 32.7 (7.2) 48.1 (13.5) 48.8 (16.8) 44.4 (15.6) < 0.01
Gender, n (%) < 0.01
Female 1493 (100%) 1209 (74.8%) 1288 (50.5%) 3990 (70.5%)
Male 408 (25.2%) 1265 (49.5%) 1673 (29.5%)
Race, n (%) < 0.01
Black 526 (20.6%) 526 (9.3%)
Hispanic 6 (0.4%) 759 (29.7%) 765 (13.5%)
White 1487 (99.6%) 1617 (100%) 1009 (39.5%) 4113 (72.6%)
Other 259 (10.1%) 259 (4.6%)
BMI (kg/m2) 26.2 (6.0) 26.8 (5.8) 28.2 (5.8) 27.3 (5.9) < 0.01
Open in a new tab

Abbreviation: BMI, body mass index.

**

p values were calculated using ANOVA for continuous variables (age and BMI) and chi‐square tests for categorical variables (gender and race) to assess differences across cohorts.

3.2. Missing Data Patterns and Imputation Outcomes

The extent of missing data varies across data sets, with GDBF exhibiting the highest proportion of missing BMD and BMI values. We found that up to 25.5% of BMD variables and 0.67% of BMI variables were missing in the GDBF study. In the NHANES study, 0.27% of BMI variables were missing. The GWAS study exhibited relatively complete phenotypic data, although additional quality control was required for some genomic markers due to low call rates.

Multiple imputations via the MICE approach were applied separately within each data set. Post‐imputation diagnostics confirmed that the imputed values closely matched the observed data distributions. Density plots comparing pre‐ and post‐imputation distributions validated the imputation model, ensuring bias introduced by missing data is minimized (Figure S2).

3.3. Assessment of Harmonization Effectiveness

The differences in BMI across the GDBF, NHANES, and GWAS data sets were statistically significant, as indicated by the Kruskal–Wallis test (p < 0.05, η 2 = 0.032), highlighting notable heterogeneity across study populations. Figure 2 compares the pre‐ and post‐harmonization distributions of BMI across the three data sets.

Figure 2.

Figure 2

Open in a new tab

Comparison of pre‐ and post‐harmonization distributions of BMI across data sets. (A) GDBF study, (B) GWAS study, and (C) NHANES cohort before harmonization; and (D) harmonized data across all cohorts, illustrating improved alignment.

Before harmonization, the BMI distribution showed notable differences, with the GDBF cohort having a more skewed distribution towards lower BMI values compared with the NHANES and GWAS data sets. Post‐harmonization, the BMI distributions across the three data sets became more aligned, reflecting the effectiveness of the harmonization process.

Post‐harmonization adjustments for age, sex, and race significantly reduced inter‐cohort variability, particularly in BMI distributions, as illustrated in Figure 3. The residual regression analyses confirmed that BMI differences across data sets decreased after harmonization. The calculated Omega squared (Ω 2 = 0.0028) indicated a meaningful reduction in the variance attributable to study differences, further supporting the success of the harmonization process. The overall BMI distribution became more aligned across data sets, demonstrating the efficacy of the harmonization process. The final data sets were well‐balanced, enabling robust downstream analyses. The racial composition across participants after harmonization demonstrates a more balanced distribution across racial groups (Figure S1).

Figure 3.

Figure 3

Open in a new tab

BMI distributions before and after harmonization adjustments. (A) Boxplots illustrating BMI values across data sets before adjustment. (B) Boxplots of BMI residuals after adjustment for age, sex, and race. (C) Density plots of BMI values before adjustment, stratified by data set. (D) Density plots of BMI residuals after adjustment for age, sex, and race, showing improved alignment across data sets.

Table S1 provides further evidence of the success of the harmonization process, highlighting effect sizes and 95% confidence intervals (CIs) for age and BMI across the GWAS, GDBF, and NHANES data sets before and after phenotype harmonization. Post‐harmonization values were more consistent across data sets, demonstrating that cohort differences were minimized while retaining essential phenotypic information.

3.4. BMD Association Analysis and Model Performance

The associations between BMD and key covariates were evaluated using GEEs and mixed‐effects models. As detailed in Table S2, race and sex remained significant predictors for the GEE model.

After harmonization, the adjusted R 2 values increased from −0.07% (GDBF) and 0.002% (GWAS) to 0.14% (harmonized data set), reflecting improved model performance and reduced inter‐study variability (Figure 4). Although this increase was relatively modest, it is important for enabling robust comparative analyses and future pooled studies. The consistency of model estimates across different statistical approaches further confirms that the harmonization process effectively preserved essential phenotype‐genotype relationships without introducing bias. In sensitivity analyses, the GEE model was re‐estimated using an independent correlation structure instead of the exchangeable structure. The results remained consistent, indicating that the model findings are robust to different correlation assumptions.

Figure 4.

Figure 4

Open in a new tab

Comparison of model estimates and residual distributions for BMD prediction. (A) Coefficient estimates with 95% confidence intervals for generalized estimating equations (GEE, red) and mixed‐effects models (blue), assessing predictors (sex, race, BMI, and age) of BMD. (B) Density plots of residuals from pre‐ and post‐harmonization models, showing consistent variance explained (R 2 values) by harmonized GEE and mixed‐effects models across data sets.

3.5. Population Structure and Genomic Harmonization Outcomes

PCA was conducted to assess genomic harmonization between the GWAS and GDBF data sets. These analyses confirmed that no systematic biases remained after harmonization. The scatter plot of PC1 and PC2 (Figure 5) illustrates substantial overlaps between the two data sets, confirming that the genomic harmonization process effectively reduced batch effects. No significant clustering by cohort was observed, suggesting that population stratification was well‐controlled in subsequent analyses.

Figure 5.

Figure 5

Open in a new tab

PCA analysis of genomic harmonization. Scatter plot of the first two principal components (PC1: 26.18%; PC2: 19.65%), illustrating substantial overlaps between GDBF and GWAS data sets, indicating effective reduction of batch effects and successful genomic harmonization.

Post‐imputation validation confirmed high genotype imputation quality, with an average R 2 of 0.987 across retained SNPs. This indicates strong concordance between imputed and expected genotype dosages.

4. Discussion

This study implemented a robust framework for harmonizing phenotypic and genomic data across three major US osteoporosis data sets: GDBF, GWAS, and NHANES. The harmonized data set addressed key challenges, including variability in phenotypic definitions, demographic imbalances, and missing data, enabling comprehensive analyses of BMD and its determinants (Doiron et al. 2013). Key findings include the successful harmonization of BMI, age, and race, reduction of inter‐study variability, and validation of known associations between BMD and predictors such as age, BMI, and race.

The harmonization framework advanced previous approaches by comprehensively addressing inconsistencies in variable definitions and implementing robust statistical techniques, including multiple imputation and residual regression analysis (White et al. 2011; Freedman 2005). This approach ensured that the harmonized data set was suitable for pooled analyses, while including diverse populations, particularly from NHANES, improved the generalizability of findings. Furthermore, genomic data were harmonized using stringent quality control and imputation protocols, enabling reliable genotype–phenotype association analyses.

This study builds upon and extends prior research in several key areas. Unlike previous studies (Wang et al. 2021; Karasik et al. 2010; Mullin et al. 2020; Richards et al. 2012; Kiel et al. 2007) that often rely on single‐cohort data sets, this work leveraged multiple data sets to enhance statistical power and representativeness. While genomic data integration has been extensively explored in other disease areas, its application to osteoporosis research remains relatively limited, primarily due to constraints in sample size (Richards et al. 2012; Graat‐Verboom et al. 2009). By harmonizing genomic and phenotypic data, we established a critical resource for investigating BMD and fracture risk across diverse populations. Additionally, the inclusion of NHANES improved demographic diversity compared with prior studies, which have predominantly focused on populations of European ancestry (Vásquez et al. 2023; Martin et al. 2019).

The findings have significant implications for both clinical practice and research. The harmonized data set lays the groundwork for developing more equitable and accurate fracture risk prediction models, which can account for population‐specific differences, addressing long‐standing disparities in osteoporosis diagnosis and treatment (Noel et al. 2020; Thomas 2007). Integrating genomic data also provides opportunities to explore gene‐environment interactions, which may reveal new biological pathways underlying BMD and fracture risk (Sambrook and Nguyen 1999). By creating a comprehensive data harmonization framework, this study provides a valuable methodological resource that can facilitate future research, including meta‐analyses and replication studies. Although the overall gain in predictive accuracy was modest (adjusted R 2 improved to 0.14 post‐harmonization) due to the limitations of the study, the reduction in inter‐cohort variability (e.g., Ω 2 decreased from 0.032 to 0.0028) indicates substantial improvement in data set balance. This suggests that model outputs became less dependent on study origin and more reflective of individual‐level characteristics. For genetic epidemiology applications, such “balance” reduces residual confounding, improves the reproducibility of SNP effect estimates, and enhances the generalizability of polygenic risk scores across cohorts. These effects were further supported by consistent model results across GEE and mixed‐effects frameworks.

This study has several limitations. The study results generated from cross‐sectional data may not be fully applicable to cohort data with phenotypic and genomic factors. Although multiple imputations effectively addressed missing data, some variables, such as alcohol consumption, were inconsistently recorded across data sets, potentially limiting the comprehensiveness of the analyses. For example, alcohol consumption was measured categorically (yes/no) in one cohort and as a continuous variable (quantity or frequency) in another, limiting full harmonization despite imputation. Additionally, while NHANES improved demographic diversity, the overrepresentation of White participants in the GDBF and GWAS data sets highlights the need for further inclusion of underrepresented populations in future research. Furthermore, while harmonization aimed to enhance model performance by reducing variability and aligning definitions across data sets, the observed improvement in predictive accuracy was relatively modest. Several factors could potentially contribute to this limited increase, including prior adjustments for key confounders (age and weight) during harmonization and the inherently complex nature of osteoporosis risk.

Although NHANES was selected for its extensive phenotypic and genomic data and its nationally representative and racially diverse sample, it was excluded from the BMD harmonization analysis due to the lack of hip BMD measurements. As a result, NHANES was used only for demographic comparisons in this study. This limits the functional contribution of NHANES to the core BMD prediction models and may reduce the generalizability of our findings to broader US populations.

While all studies used DXA, differences in scanner models and calibration may have introduced residual bias. Our use of residual regression and study‐level random effects helped mitigate this, but device‐specific calibration data were not available across all data sets. Additionally, while harmonization efforts addressed key sources of between‐study variability, some residual confounding may remain, and not all variables were perfectly aligned due to structural limitations in the original data.

Future directions include expanding the harmonization framework to longitudinal data sets and providing insights into the dynamic relationships between genomic factors and BMD over time. Efforts to include more racially and ethnically diverse populations are critical for ensuring the generalizability and equity of osteoporosis research. Additionally, incorporating other omics data, such as transcriptomics, epigenomics, proteomics, and metabolomics, could further help to elucidate the molecular mechanisms underlying osteoporosis and related conditions (Qiu et al. 2020; Zhang et al. 2021). Developing imputation‐based strategies to approximate hip BMD using alternative skeletal sites, such as those available in NHANES, could allow for more inclusive harmonization. Similarly, collecting and integrating device‐specific calibration metadata may enhance cross‐cohort comparability in future studies.

This study provides a robust and effective framework for harmonizing phenotypic and genomic data, facilitating the integration of multiple large‐scale data sets. The harmonized data enhances the accuracy and generalizability of assessments, offering potential to develop more precise and equitable fracture risk prediction tools. By addressing methodological challenges such as inconsistent variable definitions, missing data, and demographic imbalances, this work establishes a foundation for future precision medicine research in osteoporosis, particularly for populations historically underrepresented in genomic studies.

The availability of these harmonized data sets presents significant opportunities for advancing osteoporosis research, enabling more precise identification of genetic and environmental determinants of bone health. Future studies can leverage this approach to investigate diverse populations, develop targeted preventive strategies, and ultimately reduce disparities in osteoporosis diagnosis and treatment.

Author Contributions

Qing Wu: conceptualization, funding acquisition, investigation, methodology, project administration, resources, supervision, writing – original draft, writing – review and editing. Anqi Liu: data curation, formal analysis, investigation, methodology, software, validation, visualization, writing – original draft, writing – review and editing. Jianing Liu: methodology, writing – original draft, writing – review and editing. Lang Wu: methodology, writing – review and editing.

Funding

This study was supported by the National Institute on Minority Health and Health Disparities (R21MD013681, awarded to Dr. Qing Wu) and the National Institute on Aging (R01AG080017, awarded to Dr. Qing Wu).

Conflicts of Interest

Lang Wu provided consulting services to Pupil Bio Inc., Techspert, and Galiher DeRobertis & Waxman LLP, and reviewed manuscripts for Gastroenterology Report, activities unrelated to this study, and received an honorarium. The other authors declare no conflicts of interest.

Supporting information

Supplementary Table 1: Effect sizes and 95% confidence intervals (CI) for age and BMI across GWAS, GDBF, and NHANES datasets before and after phenotype harmonization. Supplementary Table 2: The estimated effects of key predictors on bone mineral density (BMD) using four different modeling approaches. Supplementary Fig 1: Proportion of racial groups among participants based on the harmonized “race” variable. Supplementary Fig 2: Density plots comparing pre‐ and post‐imputation distributions to validate the imputation model for each study cohort.

GEPI-50-0-s001.docx (229.3KB, docx)

Liu, A. , Liu J., Wu L., and Wu Q.. 2026. “Integrative Harmonization of Phenotypic and Genomic Data Improves Bone Mineral Density Prediction in Multi‐Study Osteoporosis Research.” Genetic Epidemiology 50: 1–9. 10.1002/gepi.70028.

Data Availability Statement

The data/analyses presented in the current publication are based on study data downloaded from the dbGaP website under phs000138.v2. p1 and phs000390.v1. p1, following the corresponding data use agreement. The NHANES data used in this study are publicly available from the National Center for Health Statistics (NCHS) at https://www.cdc.gov/nchs/nhanes/index.html.

References

  1. Abdi, H. , and Williams L. J.. 2010. “Principal Component Analysis.” WIREs Computational Statistics 2: 433–459. [Google Scholar]
  2. Akinbam, L. , Chen T.‐C., and Davy O., et al. 2022. National Health and Nutrition Examination Survey, 2017–March 2020 Prepandemic File: Sample Design, Estimation, and Analytic Guidelines. National Center for Health Statistics (U.S.). 10.15620/cdc:115434. [DOI] [PubMed] [Google Scholar]
  3. Albrechtsen, A. , Nielsen F. C., and Nielsen R.. 2010. “Ascertainment Biases in SNP Chips Affect Measures of Population Divergence.” Molecular Biology and Evolution 27: 2534–2547. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Azur, M. J. , Stuart E. A., Frangakis C., and Leaf P. J.. 2011. “Multiple Imputation by Chained Equations: What Is It and How Does It Work?” International Journal of Methods in Psychiatric Research 20: 40–49. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Brookes, A. J. , and Robinson P. N.. 2015. “Human Genotype–Phenotype Databases: Aims, Challenges and Opportunities.” Nature Reviews Genetics 16: 702–715. [DOI] [PubMed] [Google Scholar]
  6. Cummings, S. R. , and Melton L. J.. 2002. “Epidemiology and Outcomes of Osteoporotic Fractures.” Lancet 359: 1761–1767. [DOI] [PubMed] [Google Scholar]
  7. Doiron, D. , Burton P., Marcon Y., et al. 2013. “Data Harmonization and Federated Analysis of Population‐based Studies: The BioSHaRE Project.” Emerging Themes in Epidemiology 10: 12. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Durbin, R. 2014. “Efficient Haplotype Matching and Storage Using the Positional Burrows–Wheeler Transform (PBWT).” Bioinformatics 30: 1266–1272. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Estrada, K. , Styrkarsdottir U., Evangelou E., et al. 2012. “Genome‐Wide Meta‐Analysis Identifies 56 Bone Mineral Density Loci and Reveals 14 Loci Associated With Risk of Fracture.” Nature Genetics 44: 491–501. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Fortier, I. , Dragieva N., Saliba M., Craig C., and Robson P. J.. 2019. “Harmonization of the Health and Risk Factor Questionnaire Data of the Canadian Partnership for Tomorrow Project: A Descriptive Analysis.” CMAJ Open 7: E272–E282. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Freedman, D. 2005. Statistical Models: Theory and Practice. Cambridge University Press. 10.1017/CBO9781139165495. [DOI] [Google Scholar]
  12. Graat‐Verboom, L. , Wouters E. F. M., Smeenk F. W. J. M., van den Borne B. E. E. M., Lunde R., and Spruit M. A.. 2009. “Current Status of Research on Osteoporosis in COPD: A Systematic Review.” European Respiratory Journal 34: 209–218. [DOI] [PubMed] [Google Scholar]
  13. Hinrichs, A. L. , Culverhouse R. C., and Suarez B. K.. 2014. “Genotypic Discrepancies Arising From Imputation.” BMC Proceedings 8: S17. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Kanis, J. A. , Cooper C., Rizzoli R., and Reginster J. Y.. 2019. “European Guidance for the Diagnosis and Management of Osteoporosis in Postmenopausal Women.” Osteoporosis International 30: 3–44. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Karasik, D. , Hsu Y.‐H., Zhou Y., Cupples L. A., Kiel D. P., and Demissie S.. 2010. “Genome‐Wide Pleiotropy of Osteoporosis‐Related Phenotypes: The Framingham Study.” Journal of Bone and Mineral Research 25: 1555–1563. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Kiel, D. P. , Ferrari S. L., Cupples L. A., et al. 2007. “Genetic Variation at the Low‐Density Lipoprotein Receptor‐Related Protein 5 (LRP5) Locus Modulates Wnt Signaling and the Relationship of Physical Activity With Bone Mineral Density in Men.” Bone 40: 587–596. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Laird, N. M. , and Ware J. H.. 1982. “Random‐Effects Models for Longitudinal Data.” Biometrics 38: 963–974. [PubMed] [Google Scholar]
  18. Lakens, D. 2013. “Calculating and Reporting Effect Sizes to Facilitate Cumulative Science: A Practical Primer for T‐Tests and Anovas.” Frontiers in Psychology 4. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2013.00863. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Landry, L. G. , Ali N., Williams D. R., Rehm H. L., and Bonham V. L.. 2018. “Lack of Diversity in Genomic Databases Is a Barrier to Translating Precision Medicine Research Into Practice.” Health Affairs 37: 780–785. [DOI] [PubMed] [Google Scholar]
  20. Liang, J. 2023. Data on the Relationship Between Chloroform and Bone Mineral Density. Harvard Dataverse. 10.7910/DVN/EST0VJ. [DOI] [Google Scholar]
  21. Liang, K.‐Y. , and Zeger S. L.. 1986. “Longitudinal Data Analysis Using Generalized Linear Models.” Biometrika 73: 13–22. [Google Scholar]
  22. Little, R. , and Rubin D.. 2019. Statistical Analysis With Missing Data, Third Edition. 1st ed. Wiley. 10.1002/9781119482260. [DOI] [Google Scholar]
  23. Luningham, J. M. , McArtor D. B., Hendriks A. M., et al. 2019. “Data Integration Methods for Phenotype Harmonization in Multi‐Cohort Genome‐Wide Association Studies With Behavioral Outcomes.” Frontiers in Genetics 10: 1227. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Martin, A. R. , Kanai M., Kamatani Y., Okada Y., Neale B. M., and Daly M. J.. 2019. “Clinical Use of Current Polygenic Risk Scores May Exacerbate Health Disparities.” Nature Genetics 51: 584–591. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Mullin, B. H. , Tickner J., Zhu K., et al. 2020. “Characterisation of Genetic Regulatory Effects for Osteoporosis Risk Variants in Human Osteoclasts.” Genome Biology 21: 80. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. National Center for Biotechnology Information . n.d‐a. Genomic Wide Scans for Female Osteoporosis Genes. https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000390.v1.p1.
  27. National Center for Biotechnology Information . n.d‐b. GWAS for Genetic Determinants of Bone Fragility in European‐American Premenopausal Women. https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000138.v2.p1.
  28. Noel, S. E. , Santos M. P., and Wright N. C.. 2020. “Racial and Ethnic Disparities in Bone Health and Outcomes in the United States.” Journal of Bone and Mineral Research 36: 1881–1905. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Pan, K. , Bazzano L. A., Betha K., et al. 2023. “Large‐Scale Data Harmonization Across Prospective Studies.” American Journal of Epidemiology 192: 2033–2049. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Popejoy, A. B. , and Fullerton S. M.. 2016. “Genomics Is Failing on Diversity.” Nature 538: 161–164. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Price, A. L. , Patterson N. J., Plenge R. M., Weinblatt M. E., Shadick N. A., and Reich D.. 2006. “Principal Components Analysis Corrects for Stratification in Genome‐Wide Association Studies.” Nature Genetics 38: 904–909. [DOI] [PubMed] [Google Scholar]
  32. Qiu, C. , Yu F., Su K., et al. 2020. “Multi‐Omics Data Integration for Identifying Osteoporosis Biomarkers and Their Biological Interaction and Causal Mechanisms.” iScience 23: 100847. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Richards, J. B. , Zheng H.‐F., and Spector T. D.. 2012. “Genetics of Osteoporosis From Genome‐Wide Association Studies: Advances and Challenges.” Nature Reviews Genetics 13: 576–588. [DOI] [PubMed] [Google Scholar]
  34. Sambrook, P. , and Nguyen T.. 1999. “Bone Mineral Density and Gene‐Environment Interactions in the Search for Osteoporosis Genes.” Environmental Health Perspectives 107, no. 3: A130–1. 10.1289/ehp.107-1566373. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Schmidt, C. O. , Struckmann S., Enzenbach C., et al. 2021. “Facilitating Harmonized Data Quality Assessments. A Data Quality Framework for Observational Health Research Data Collections With Software Implementations in R.” BMC Medical Research Methodology 21: 63. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Singer, A. J. , Sharma A., Deignan C., and Borgermans L.. 2023. “Closing the Gap in Osteoporosis Management: The Critical Role of Primary Care in Bone Health.” Current Medical Research and Opinion 39: 387–398. [DOI] [PubMed] [Google Scholar]
  37. Sozen, T. , Ozisik L., and Calik Basaran N.. 2017. “An Overview and Management of Osteoporosis.” European Journal of Rheumatology 4: 46–56. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Stilp, A. M. , Emery L. S., Broome J. G., et al. 2021. “A System for Phenotype Harmonization in the National Heart, Lung, and Blood Institute Trans‐Omics for Precision Medicine (TOPMed) Program.” American Journal of Epidemiology 190: 1977–1992. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Sung, Y. J. , De Las Fuentes L., Winkler T. W., et al. 2019. “A Multi‐Ancestry Genome‐Wide Study Incorporating Gene–Smoking Interactions Identifies Multiple New Loci for Pulse Pressure and Mean Arterial Pressure.” Human Molecular Genetics 28: 2615–2633. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. The Haplotype Reference Consortium . 2016. “A Reference Panel of 64,976 Haplotypes for Genotype Imputation.” Nature Genetics 48: 1279–1283. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Thomas, P. A. 2007. “Racial and Ethnic Differences in Osteoporosis.” Journal of the American Academy of Orthopaedic Surgeons 15: S26–S30. [DOI] [PubMed] [Google Scholar]
  42. Vásquez, E. , Alam M. T., and Murillo R.. 2023. “Race and Ethnic Differences in Physical Activity, Osteopenia, and Osteoporosis: Results From NHANES 2009–2010, 2013–2014, 2017–2018.” Archives of Osteoporosis 19: 7. [DOI] [PubMed] [Google Scholar]
  43. Wang, L. , Yu W., Yin X., et al. 2021. “Prevalence of Osteoporosis and Fracture in China: The China Osteoporosis Prevalence Study.” JAMA Network Open 4: e2121106. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. White, I. R. , Royston P., and Wood A. M.. 2011. “Multiple Imputation Using Chained Equations: Issues and Guidance for Practice.” Statistics in Medicine 30: 377–399. [DOI] [PubMed] [Google Scholar]
  45. Wong, M. , Day N., Luan J., Chan K., and Wareham N.. 2003. “The Detection of Gene–Environment Interaction for Continuous Traits: Should We Deal With Measurement Error by Bigger Studies or Better Measurement?” International Journal of Epidemiology 32: 51–57. [DOI] [PubMed] [Google Scholar]
  46. Yuan, J. , Tickner J., Mullin B. H., et al. 2019. “Advanced Genetic Approaches in Discovery and Characterization of Genes Involved With Osteoporosis in Mouse and Human.” Frontiers in Genetics 10: 288. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Zhang, X. , Xu H., Li G. H., et al. 2021. “Metabolomics Insights into Osteoporosis Through Association With Bone Mineral Density.” Journal of Bone and Mineral Research 36: 729–738. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Zheng, H. F. , Forgetta V., Hsu Y. H., et al. 2015. “Whole‐Genome Sequencing Identifies EN1 as a Determinant of Bone Density and Fracture.” Nature 526: 112–117. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Table 1: Effect sizes and 95% confidence intervals (CI) for age and BMI across GWAS, GDBF, and NHANES datasets before and after phenotype harmonization. Supplementary Table 2: The estimated effects of key predictors on bone mineral density (BMD) using four different modeling approaches. Supplementary Fig 1: Proportion of racial groups among participants based on the harmonized “race” variable. Supplementary Fig 2: Density plots comparing pre‐ and post‐imputation distributions to validate the imputation model for each study cohort.

GEPI-50-0-s001.docx (229.3KB, docx)

Data Availability Statement

The data/analyses presented in the current publication are based on study data downloaded from the dbGaP website under phs000138.v2. p1 and phs000390.v1. p1, following the corresponding data use agreement. The NHANES data used in this study are publicly available from the National Center for Health Statistics (NCHS) at https://www.cdc.gov/nchs/nhanes/index.html.

Articles from Genetic Epidemiology are provided here courtesy of Wiley

RESOURCES