Incorporating Polygenic Risk Scores and Nongenetic Risk Factors for Breast Cancer Risk Prediction Among Asian Women

Key Points

Question  How well do breast cancer risk prediction models that incorporate polygenic risk scores (PRSs) and nongenetic risk factors perform for Asian women?

Findings  In this diagnostic study of 126 894 women, a PRS including 111 genetic variants was developed and tested using data from a prospective cohort study. The PRS was significantly associated with breast cancer risk, and adding 7 nongenetic risk factors improved the model’s accuracy.

Meaning  These findings support the utility of prediction models in identifying Asian women with high risk of breast cancer.

Importance  Polygenic risk scores (PRSs) have shown promise in breast cancer risk prediction; however, limited studies have been conducted among Asian women.

Objective  To develop breast cancer risk prediction models for Asian women incorporating PRSs and nongenetic risk factors.

Design, Setting, and Participants  This diagnostic study included women of Asian ancestry from the Asia Breast Cancer Consortium. PRSs were developed using data from genomewide association studies (GWASs) of breast cancer conducted among 123 041 women with Asian ancestry (including 18 650 women with breast cancer) using 3 approaches: (1) reported PRS for women with European ancestry; (2) breast cancer–associated single-nucleotide variations (SNVs) identified by fine-mapping of GWAS-identified risk loci; and (3) genomewide risk prediction algorithms. A nongenetic risk score (NGRS) was built, including 7 well-established nongenetic risk factors, using data of 416 case participants and 1558 control participants from a prospective cohort study. PRSs were initially validated in an independent data set including 1426 case participants and 1323 control participants and further evaluated, along with the NGRS, in the second data set including 368 case participants and 736 control participants nested within a prospective cohort study.

Main Outcomes and Measures  Logistic regression was used to examine associations of risk scores with breast cancer risk to estimate odds ratios (ORs) with 95% CIs and area under the receiver operating characteristic curve (AUC).

Results  A total of 126 894 women of Asian ancestry were included; 20 444 (16.1%) had breast cancer. The mean (SD) age ranged from 49.1 (10.8) to 54.4 (10.4) years for case participants and 50.6 (9.5) to 54.0 (7.4) years for control participants among studies that provided demographic characteristics. In the prospective cohort, a PRS with 111 SNVs developed using the fine-mapping approach (PRS111) showed a prediction performance comparable with a genomewide PRS that included more than 855 000 SNVs. The OR per SD increase of PRS111 score was 1.67 (95% CI, 1.46-1.92), with an AUC of 0.639 (95% CI, 0.604-0.674). The NGRS had a limited predictive ability (AUC, 0.565; 95% CI, 0.529-0.601). Compared with the average risk group (40th-60th percentile), women in the top 5% of PRS111 and NGRS were at a 3.84-fold (95% CI, 2.30-6.46) and 2.10-fold (95% CI, 1.22-3.62) higher risk of breast cancer, respectively. The prediction model including both PRS111 and NGRS achieved the highest prediction accuracy (AUC, 0.648; 95% CI, 0.613-0.682).

Conclusions and Relevance  In this study, PRSs derived using breast cancer risk–associated SNVs had similar predictive performance in Asian and European women. Including nongenetic risk factors in models further improved prediction accuracy. These findings support the utility of these models in developing personalized screening and prevention strategies.

Breast cancer is the most commonly diagnosed malignant neoplasm among women worldwide.¹ The incident rate of breast cancer has been increasing substantially in multiple Asian countries, many of which do not have a population-based breast cancer screening program, leading to delayed cancer diagnoses and poor survival rates.² Because of the differences in breast cancer risk, screening programs currently implemented in the US and European countries may not be appropriate for Asian countries. Hence, a cost-efficient, population-specific breast cancer screening strategy for Asian women is needed.

In 2006, we established the Asia Breast Cancer Consortium (ABCC) to identify genetic risk variants, including single nucleotide variants (SNVs), associated with breast cancer risk through genomewide association studies (GWASs). Approximately 50 risk loci were identified using Asian data alone or in combination with data from European descendants.^3-10 However, most breast cancer risk loci were identified in GWASs conducted in European descendants.¹¹ Multiple studies have attempted to aggregate the associations of risk SNVs identified by GWAS as polygenic risk scores (PRSs) to stratify women into different breast cancer risk groups.^12-16 Most PRSs for breast cancer were established specifically in women of European ancestry, and among them, a 313-SNV PRS showed the highest predictive ability, with an area under the receiver operating characteristic curve (AUC) of 0.630 to 0.642.¹² To our knowledge, few studies of breast cancer PRSs for Asian women have been conducted, and limited prediction accuracy was observed.^17-21 A recent study showed that the 313-SNV PRS performed better than PRSs derived from Asian data.²¹ However, in that study, the Asian-specific PRSs were derived using a limited number of SNVs.

Nongenetic factors are also associated with breast cancer risk.²² Several studies have explored the potential of incorporating PRSs and nongenetic risk factors in improving predictive accuracy.^16,22-25 Among them, a recent study among European women revealed that the 313-SNV PRS was more predictive than a model including 16 nongenetic risk factors, and the best risk stratification performance was achieved when PRS and nongenetic factors were combined.²² However, to our knowledge, similar studies have rarely been carried out among Asian women. In the present study, we aimed to develop and validate PRSs for Asian women using data from the largest GWAS of breast cancer conducted among Asian women that we are aware of and to further evaluate the performance of risk prediction models, including both PRSs and nongenetic risk factors.

In this diagnostic study, the PRS development data sets included GWAS data of 20 076 women with breast cancer and 105 714 control participants of Asian ancestry from the ABCC (Table 1). Detailed information on the ABCC is described elsewhere.¹⁰ We divided these data sets to a training set (18 650 case participants and 104 391 control participants) for PRS derivation and a validation set (1426 case participants and 1323 control participants) for prediction performance evaluation (eMethods in the Supplement). For each PRS development approach, the most predictive PRS in our validation set was further evaluated in an independent case-control study nested within a prospective cohort study, comprising 368 case participants and 736 control participants, individually matched by age (<5 years). Included in the nested case-control study were participants from the Shanghai Women’s Health Study (SWHS), and none of them had a diagnosis of any cancers at the time of enrollment (eMethods in the Supplement).^10,26 All studies involved in the current analyses have been approved by their respective institutional review boards. All participants provided informed consent prior to study inclusion. This study followed the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) reporting guideline.

Detailed descriptions of genetic data are provided in our recent publication¹⁰ and in the eMethods in the Supplement. Genotyping was conducted using several platforms, and genotyping data imputation was performed separately by study (eTable 1 in the Supplement). GWAS was conducted within each study and/or substudy, and association results were combined via fixed-effects meta-analyses.

We applied 3 approaches to develop PRSs, as described briefly here and in detail in the eMethods in the Supplement. PRSs were calculated using the following formula:

in which SNV_k and β_k represent the allelic dosage and corresponding weight of SNV k, and n is the number of SNVs used.

To our knowledge, the best breast cancer PRS to date was the 313-SNV PRS among women with European ancestry.¹² Most recently, this PRS was updated by adding 17 novel breast cancer–susceptibility SNVs.¹¹ Of these 330 SNVs, 263 could be found in our validation set and prospective test set, and 3 PRSs (PRS263-Asian, PRS263-European, and PRS263-meta) were derived using weights of these SNVs from data from women with European ancestry included in the Breast Cancer Association Consortium (BCAC-European),¹¹ data from women with Asian ancestry in our training set (Table 1), and meta-analyses of these 2 data sets, respectively.

As shown in the eFigure in the Supplement, for each of the 238 independent susceptibility loci for breast cancer,^3-10,27 fine-mapping analyses were performed using data from our training set to identify SNVs that were independently associated with breast cancer using Genomewide Complex Trait Analysis–conditional and joint association analysis (GCTA-COJO) version 1.93.2.^28,29 Within each locus, a COJO threshold of P < 1 × 10⁻⁵ was used to identify independently associated SNVs and reestimate weights of their association with breast cancer risk. Some loci were ineligible for fine-mapping because no SNVs within them had an association with breast cancer risk at P < 10⁻⁵ in our training set. Based on fine-mapping, 3 PRSs were derived using (1) all SNVs selected from fine-mapping; (2) SNVs selected by fine-mapping and showing consistent association directions with P < .05 in the BCAC-European data; and (3) SNVs in the second PRS and lead SNVs from loci that were ineligible for fine-mapping but showed P < .05 in our training set (eFigure in the Supplement). We repeated the fine-mapping analyses at COJO thresholds of P < 1 × 10⁻³ and P < 1 × 10⁻⁴ to identify SNVs and used them to construct 3 sets of PRSs for each threshold following the same steps described previously.

LDpred (version 1.0.11), LDpred2 (version 1.4.4), and PRS-CSx (July 29, 2021, release) were used to derive PRSs using data from the training set. Details of these algorithms can be obtained elsewhere.^30-32 Of a total of 5 947 015 SNVs, indels and ambiguous SNVs were excluded by LDpred, and weights of the associations of the remaining 4 487 284 SNVs with breast cancer were reevaluated. Both LDpred2 and PRS-CSx recommend using SNVs included in HapMap 3; thus, of the 5 947 015 SNVs, weights of only 855 680 HapMap 3 SNVs were reestimated (eMethods in the Supplement).

Established nongenetic breast cancer risk factors included body mass index (BMI; calculated as weight in kilograms divided by height in meters squared), waist-to-hip ratio (WHR), benign breast disease, age at menarche, age at first live birth, and family history of breast cancer. The interaction between BMI and menopause status was included in the model, as BMI shows a different association with breast cancer risk by menopausal status.¹⁷ Data from 1974 women (416 case participants and 1558 control participants) from the SWHS, but independent from those in the prospective test set, were used to estimate the weights of the association of these 7 nongenetic factors and the interaction term with breast cancer risk (eTable 2 in the Supplement). Missing data were only found for age at menarche (1 and 2 participants in the training and test sets, respectively) and age at first live birth (72 and 40 participants in the training and test sets, respectively), which were imputed using the R function mice::mice (R version 3.6.0 [R Project for Statistical Computing]). A logistic regression model was fitted with case or control status as the outcome and these 8 factors set as predictors. Weights estimated from this model were then used to construct a nongenetic risk score (NGRS) for each participant using the following formula:

where F_k and w_k are the value and corresponding weight of factor k, and w_i is the weight of the interaction term between BMI and menopause status.

For the PRS showing the highest prediction accuracy in our prospective test set and the reported European PRS,¹² an integrated risk prediction model (IRPM) was built including a PRS and the NGRS to predict breast cancer risk (breast cancer risk equals approximately PRS + NGRS). Logistic regression was used to evaluate ORs and 95% CIs per SD increase in these risk scores. Prediction performance was measured by AUCs and 95% CIs using the R function pROC:roc.³³

We estimated the 10-year absolute risk of developing breast cancer using the most predictive PRS in our prospective test set and the reported European PRS.¹² A total of 10 207 Chinese women (5087 case participants and 5120 control participants) from the Shanghai Breast Cancer Genetic Study and BCAC-Asians (eTable 1 in the Supplement) were included in this analysis. Logistic regression was used to estimate breast cancer ORs of different PRS percentile groups compared with the middle quintile (40th to 60th percentile) group. Then 10-year absolute risks were calculated using these ORs and the incidence and mortality rates of breast cancer in Shanghai in 2017 following the strategy described previously.¹⁷ ORs with a 2-tailed P < .05 were considered statistically significant.

A total of 126 894 women of Asian ancestry from the ABCC, including 20 444 (16.1%) with breast cancer and 106 450 control participants, were included in this study (Table 1). The ABCC data sets were divided into a PRS derivation set (18 650 case participants and 104 391 control participants), a validation set (1426 case participants and 1323 control participants), and a prospective test set (368 case participants and 736 control participants). The mean (SD) age ranged from 49.1 (10.8) to 54.4 (10.4) years for case participants and 50.6 (9.5) to 54.0 (7.4) years for control participants among studies that provided demographic characteristics data to the consortium.

In general, all PRSs had a slightly higher mean value among case participants than among control participants, while the SD value was similar between both groups, in both the case-control validation set and the prospective test set (eTable 3 in the Supplement). The 3 PRSs derived based on the reported European PRS had similar prediction performance in our validation set (Table 2 and eTable 3 and eTable 4 in the Supplement). However, in our prospective test set, PRS263-meta had the best prediction accuracy (AUC, 0.626; 95% CI, 0.592-0.661) (Table 2). The OR of breast cancer per SD increase of PRS263-meta score was 1.63 (95% CI, 1.43-1.87).

Using the fine-mapping approach, 3 PRSs were developed at each fine-mapping threshold (eTable 3 in the Supplement), and among them, PRS111 showed the strongest association with breast cancer risk and highest prediction performance in both validation and prospective test sets (Table 2). This PRS was developed using 57 SNVs selected by fine-mapping and showing consistent association directions with P < .05 in the BCAC-European data,¹¹ plus 54 lead SNVs in GWAS loci with P < .05 in our training set (eFigure and eTable 5 in the Supplement). The OR for breast cancer per SD increase in PRS111 score was 1.45 (95% CI, 1.34-1.57) in our validation set and 1.67 (95% CI, 1.46-1.92) in our prospective test set, with AUCs of 0.603 (95% CI, 0.582-0.624) and 0.639 (95% CI, 0.604-0.674), respectively (Table 2). Compared with the average risk group (40th-60th percentile), women in the top 5% of PRS111 were at 3.84-fold (95% CI, 2.30-6.46) increased risk of breast cancer. Among participants younger than 60 years, the OR of the association between PRS111 score and breast cancer risk increased with age; however, this reversed among women older than 60 years (eTable 6 in the Supplement). No significant interaction between PRS111 and age was observed. For both PRS111 and PRS263-meta, distribution curves for case participants were shifted to the right compared with those for control participants, and the overlap was less for PRS111 than PRS263-meta (Figure 1)_. The difference in median percentile between case and control participants was higher for PRS111 (64 [37-85] vs 43 [21-69]) compared with PRS263-meta (60 [37-83] vs 44 [20-71]) (Figure 1).

Among the most predictive PRS derived by each genomewide risk prediction algorithm, the PRS derived from PRS-CSx was the best-performing PRS in both the validation and the prospective test set, with AUCs of 0.613 (95% CI, 0.592-0.634) and 0.642 (95% CI, 0.608-0.676), respectively (Table 2 and eTable 3 in the Supplement). The OR for breast cancer per SD increase of this PRS in these 3 data sets was 1.51 (95% CI, 1.39-1.63) and 1.70 (95% CI, 1.49-1.95), respectively. Although the PRS derived from PRS-CSx performed slightly better than PRS111 (AUCs, 0.642 vs 0.639), we chose PRS111 for downstream analyses because it used fewer SNVs than the PRS derived from PRS-CSx (111 vs 855 680 SNVs) but had almost equal predictive ability.

In the training set for NGRS, compared with control participants, case participants, on average, had a younger age at menarche, older age at first live birth, and higher BMI (eTable 2 in the Supplement). More case than control participants were postmenopausal and had a family history of breast cancer or a prior diagnosis of benign breast diseases. Although not all statistically significant, these findings were consistent with those that have been reported previously. Similar patterns of the associations were found in the prospective test set, although not all associations were statistically significant. Compared with the training set, women included in the prospective test set tended to be older, were more likely to be postmenopausal (576 of 1104 [52.2%] vs 908 of 1974 [46.0%]), and had a lower prevalence of prior benign breast disease diagnosis (174 of 1104 [15.8%] vs 395 of 1974 [20.0%]).

In our prospective test set, the NGRS was associated with breast cancer risk with an OR per SD increase of 1.29 (95% CI, 1.14-1.46) and an AUC of 0.565 (95% CI, 0.529-0.601) (Table 3). Incorporating the NGRS with PRS111 or PRS263-meta, we established 2 models, IRPM111 and IRPM263-meta, respectively. Within both models, the PRS and the NGRS were independently associated with breast cancer, although the PRS showed a stronger association than the NGRS (Table 3). IRPM111 showed a better prediction accuracy (AUC, 0.648; 95% CI, 0.613-0.682) than IRPM263-meta (AUC, 0.632; 95% CI, 0.597-0.666) (Table 3). In the prospective test set, family history was associated with breast cancer risk independently of other nongenetic factors (OR, 2.17; 95% CI, 1.10-4.28), and this association was slightly attenuated after adjusting for PRS111 (adjusted OR, 2.11; 95% CI, 1.06-4.22).

Among the 10 207 Chinese women from the ABCC data sets, a dose-response association of breast cancer risk with percentiles of PRS111 and PRS263-meta was observed (Figure 2A and B). Compared with the average risk group, women in the top 5% of PRS111 and PRS263-meta were at a 3.39-fold (95% CI, 2.80-4.10) and 2.23-fold (95% CI, 1.87-2.65) increased risk of breast cancer, respectively; while those at the bottom 5% were at 0.30-fold (95% CI, 0.24-0.39) and 0.44-fold (95% CI, 0.35-0.56) decreased risk of breast cancer, respectively (eTable 7 in the Supplement). The 10-year absolute risks were estimated by PRS111 and PRS263-meta percentiles and age groups. For women aged 60 years, the ranges of 10-year absolute risks estimated by PRS111 and PRS263-meta were 0.35% to 7.68% and 0.73% to 5.32%, respectively (Figure 2C and D)

Leveraging large GWAS data sets of women with Asian ancestry, we developed PRSs for breast cancer risk using 3 approaches and validated their prediction performance in an independent prospective test cohort. PRS111, derived using the fine-mapping approach, was the best-performing PRS in this study (AUC, 0.639). The prediction model incorporating PRS111 and 7 nongenetic risk factors achieved a further improved prediction accuracy (AUC, 0.648).

A recent study compared the predictive ability of 5 Asian-specific PRSs with that of the 313-SNV European PRS in Asian women.²¹ The 313-SNV PRS was significantly more predictive (AUC, 0.617) than any of the 5 Asian-specific PRSs (best AUC, 0.586).²¹ However, because most of the breast cancer risk variants were identified in GWAS conducted among women with European ancestry, the Asian-specific PRSs were derived using a limited number of SNVs (from 5 to 51) in that study. In the present study, the most predictive PRS based on these 313 SNVs, PRS263-meta, underperformed the PRS111, which was derived entirely using Asian data. The prediction ability of PRS111 in Asian women (AUC, 0.639) is almost equivalent with that of the 313-SNV PRS in women with European ancestry (AUC, 0.642).¹² The prediction ability of our PRS111 is also comparable with that derived using genomewide data in a recent study conducted among Asian women (AUC, 0.635).³⁴

Most studies of prediction models incorporating PRS and nongenetic risk factors were carried out among European descendants.^16,24,35-38 Generally, including nongenetic risk factors could lead to improved prediction accuracy, although the magnitude of improvement is relatively small. In a recent analysis using data from a prospective cohort of Dutch women, the 313-SNV European PRS was found to have an AUC of 0.636.³⁷ Incorporating this PRS with 9 nongenetic risk factors improved the AUC to 0.653,³⁷ similar to the level achieved in our study. In 2010, we built an Asian-specific prediction model incorporating a 12-SNV PRS and 5 nongenetic risk factors, which showed an AUC of 0.629 among Chinese women.¹⁷ In the present study, IRPM111, the prediction model including PRS111 and the NGRS, outperformed both PRS111 and the NGRS in predicting breast cancer risk.

The strengths of this study include the use of large GWAS data sets as the training set to improve the accuracy of estimating weights of breast cancer–associated SNVs for PRS construction. We performed fine-mapping analyses to identify additional breast cancer risk SNVs specifically for Asian women. Because most of the breast cancer–associated SNVs were identified in populations with European ancestry and there are differences in genetic architectures between Asian and European populations, we believe that this approach is necessary to construct a PRS that is more appropriate for Asian women. In addition, we demonstrated the ability of PRS-CSx in developing more predictive PRSs than other genomewide prediction algorithms. Finally, the availability of both genetic and nongenetic risk factors data made it possible to establish and validate prediction models incorporating PRSs and nongenetic risk factors.

This study has limitations. The sample size of our prospective test set was relatively small, which led to relatively wide 95% CIs for ORs and AUCs. Although the PRSs used for relative risk estimation were externally validated, there might still be some potential for overfitting in risk estimation. We found that all PRSs had better prediction performance in our prospective test set than in our case-control validation set. Reasons for this observation are unclear. Case and control participants were from 2 different studies in the case-control validation set, which could reduce the comparability between the case and control groups. However, the patterns of associations uncovered in both validation tests were similar. Additionally, including additional nongenetic factors, such as mammographic density, could improve model performance; however, such data were not available in the present study.

Using data from the largest GWAS conducted in Asian women, we demonstrated that PRSs derived using breast cancer–associated risk SNVs show similar performance in predicting breast cancer risk in Asian and European descendants. Including known nongenetic risk factors in the models could further improve the accuracy of risk prediction. Our study provides strong support for the utility of risk prediction models in developing personalized screening and prevention strategies.

Accepted for Publication: December 23, 2021.

Published: March 21, 2022. doi:10.1001/jamanetworkopen.2021.49030

Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2022 Yang Y et al. JAMA Network Open.

Corresponding Author: Wei Zheng, MD, PhD, Division of Epidemiology, Department of Medicine, Vanderbilt Epidemiology Center, Vanderbilt-Ingram Cancer Center, Vanderbilt University Medical Center, 2525 W End Ave, Ste 800, Nashville, TN 37203 (wei.zheng@vanderbilt.edu).

Author Contributions: Dr W. Zheng had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: W. Zheng.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Yang, Kong, B. Park, W. Zheng.

Critical revision of the manuscript for important intellectual content: Yang, Tao, X. Shu, Cai, Wen, Gu, Gao, Y. Zheng, Kweon, Shin, Choi, Lee, M. Park, Jia, Li, Kang, X.-O. Shu, Long, W. Zheng.

Statistical analysis: Yang, Tao, Wen, Gu, Jia, Li.

Obtained funding: W. Zheng.

Administrative, technical, or material support: Cai, Gu, Y. Zheng, Kong, B. Park, W. Zheng.

Supervision: Gao, Lee, M. Park, Kang, Long, W. Zheng.

Conflict of Interest Disclosures: None reported.

Funding/Support: This research was supported in part by grants R01CA235553, R01CA124558, R01CA158473, and R01CA148667 from the National Institutes of Health. Sample preparation and genotyping assays at Vanderbilt were conducted at the Survey and Biospecimen Shared Resources and Vanderbilt Microarray Shared Resource, which are supported in part by the Vanderbilt-Ingram Cancer Center (grant No. P30CA068485). Data analyses were conducted using the Advanced Computing Center for Research and Education at Vanderbilt University. The Seoul Breast Cancer Study was supported by the Basic Research Laboratory program through the National Research Foundation of Korea, funded by the Ministry of Education, Science, and Technology (grant No. 2011-0001564). The Korean Hereditary Breast Cancer Study/Korean Genome Epidemiology Study was supported by a grant from the National Research & Development Program for Cancer Control, Ministry for Health, Welfare, and Family Affairs, Republic of Korea (grant No. 1020350). The Korean Precision Oncology Program was supported by grant-in-aid for cancer research and control from the National Cancer Center of Korea (grant Nos. 1410690 and 1710170). Studies conducted among Asian women include the Shanghai Breast Cancer Study (principal investigators, Dr W. Zheng and Dr X.-O. Shu; grant No. R01CA064277), the Shanghai Women’s Health Study (principal investigator, Dr W. Zheng; grant Nos. R37CA070867 and UM1CA182910), the Shanghai Breast Cancer Survival Study (principal investigator, Dr X.-O. Shu; grant No. R01CA118229), the Shanghai Endometrial Cancer Study (principal investigator, Dr X.-O. Shu; grant No. R01CA092585; data for control participants only used in this study), the Seoul Breast Cancer Study (principal investigator, Dr Kang; Basic Research Laboratory program through the National Research Foundation of Korea, funded by the Ministry of Education, Science, and Technology grant 2012-0000347), the BioBank Japan Project (principal investigator, Michiaki Kubo; the Ministry of Education, Culture, Sports, Sciences and Technology from the Japanese Government); the Hwasun Cancer Epidemiology Study-Breast (Dr Kweon; the Biobank of Chonnam National University Hwasun Hospital, a member of the Korea Biobank Network, grant No. 07SA2014020). The Breast Cancer Association Consortium is funded by Cancer Research UK (grant No. C1287/A16563), the European Community’s Seventh Framework Programme (grant No. 223175 [HEALTH-F2-2009-223175] COGS).

Role of the Funder/Sponsor: The funders had no role in design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Disclaimer: The content is solely the responsibility of the authors and does not necessarily represent the official views of the funding agents.