Improving preventive screening efficiency: A population-based model of age-specific mammographic density for breast Cancer detection in Saudi Arabia

Abstract

Objective

Current age-based breast cancer screening protocols may not be optimally effective as they overlook mammographic density as a key risk factor. This study developed a personalized risk stratification model by analyzing age-specific mammographic density patterns to improve screening accuracy and reduce false-positive rates.

Methods

A cross-sectional analysis was performed on mammographic data from 2584 women aged 32–90 years from October 2023–December 2024. Breast Imaging Reporting and Data System (BI-RADS) density classifications were analyzed using polynomial regression and changepoint analysis to identify critical age thresholds. Four age-density clusters were derived, and a gradient boosting model was developed to evaluate predictive accuracy.

Results

The analysis identified three significant age thresholds (42.3, 51.7, and 65.2 years) where mammographic density patterns shifted. Four risk clusters were established, and the model achieved high predictive accuracy (Area Under the Curve [AUC] = 0.83). Simulations projected that personalized screening protocols could increase cancer detection by 14.7 % and reduce false positives by 9.7 % compared to traditional age-only approaches.

Conclusions

Age-specific mammographic density screening offers a data-driven method to advance breast cancer prevention. It provides a framework for developing more effective screening policies that can decrease morbidity, supporting a shift toward risk-based screening as standard care.

Highlights

• •Breast density changes significantly at three key age thresholds.
• •A new model can personalize breast cancer risk prediction.
• •Personalized screening can boost cancer detection by 14.7 %.
• •Personalized screening can cut false positives by 9.7 %.

1. Introduction

Breast cancer remains a leading cause of cancer mortality among women worldwide, with early detection being crucial for improved outcomes. Traditional screening approaches rely primarily on age-based guidelines, applying the same screening protocol to all women within specific age groups. However, this one-size-fits-all approach fails to account for individual variations in risk factors, potentially leading to both missed cancers in high-risk women and unnecessary procedures in low-risk women (Walker et al., 2024). As Walker and colleagues note, the current practice of screening based solely on age fails to account for the variation in breast cancer risk across the population, leading to inappropriate screening levels for many individuals. This limitation has prompted considerable international attention on moving toward more personalized screening strategies that incorporate individual risk assessment (Walker et al., 2024).

Mammographic breast density is a significant independent risk factor for breast cancer, with women having extremely dense tissue showing three to six times higher risk than those with fatty tissue (Knerr et al., 2017). High density also reduces mammographic sensitivity, creating a “double jeopardy” where high-risk women are more likely to have missed cancers (Destounis et al., 2020).

Breast density typically decreases with age, but patterns vary significantly among women (Loewen et al., 2022). Louro and colleagues (Louro et al., 2023) demonstrated this variability in 57,411 women, finding 4-year breast cancer risk ranging from 0.22 % to 7.33 %. These substantial risk differences support personalized screening protocols over uniform age-based approaches (Zhu et al., 2024).

Recent advances have made personalized screening increasingly feasible. The Personalized Risk Assessment for Prevention and Early Detection of Breast Cancer: Integration and Implementation project incorporated breast density, family history, and genetic information (Walker et al., 2024), while the Risk-based Breast Screening study explored risk-based screening with digital breast tomosynthesis (Caumo et al., 2025). These represent a growing trend toward risk-stratified approaches optimizing resource allocation and detection rates. However, significant gaps remain. Most models use simplistic density categorizations rather than detailed age-specific patterns. Few studies examine how density distributions vary across age groups or translate risk models into implementable protocols (Freer, 2015). Resource optimization is often overlooked despite its importance for practical implementation (Guerrero-Nancuante et al., 2025).

Our study addresses these gaps with 2584 women aged 32–90 years (mean 53.98) and standardized density classifications. By analyzing age-density relationships, we aim to develop a risk stratification model informing personalized screening protocols that improve early detection while minimizing unnecessary procedures.

2. Methods

2.1. Study design and population

This retrospective cross-sectional study analyzed data from 2584 women who participated in a breast cancer screening campaign between October 2023 and December 2024. The study protocol was approved by the Taif health cluster ethics committee, [2025-E-40]. Participants ranged in age from 32 to 90 years (mean age: 53.98 years, median: 53.0 years) and were categorized into five age groups: <40 (n = 35, 1.4 %), 40–49 (n = 1019, 39.4 %), 50–59 (n = 908, 35.1 %), 60–69 (n = 525, 20.3 %), and 70+ (n = 96, 3.7 %). All participants underwent standard mammographic screening with breast density assessment according to the Breast Imaging Reporting and Data System (BI-RADS) (Kerlikowske et al., 2022).

2.2. Data collection

Demographic information and mammographic data were collected for all participants on May and June 2025 from two major healthcare centers in Saudi Arabia: King Abdulaziz Hospital and King Faisal Medical Complex. Breast density was classified according to the standardized BI-RADS categories: A (predominantly fatty, n = 588, 22.8 %), B (scattered fibroglandular densities, n = 1365, 52.8 %), C (heterogeneously dense, n = 528, 20.4 %), and D (extremely dense, n = 102, 3.9 %). One participant had missing breast density data and was excluded from specific analyses requiring this variable. Mammographic assessments were performed by experienced radiologists who were blinded to the study objectives. To ensure consistency, a random sample of 10 % of mammograms was independently reviewed by a second breast imaging consultant radiologist, with an inter-rater reliability of κ = 0.82 (95 % confidence interval [CI], 0.78–0.86) (Berg et al., 2012).

2.3. Statistical analysis

We performed descriptive analyses to characterize the distribution of breast density across the entire population and within each age group. Chi-square tests were used to assess associations between categorical variables, and analysis of variance (ANOVA) was employed to compare continuous variables across groups. All statistical tests were two-sided, with a significance level of α = 0.05.

To identify patterns in the relationship between age and breast density, we employed multiple analytical approaches. First, we used polynomial regression models of varying degrees (1–4) to characterize the non-linear relationship between age (continuous variable) and breast density (ordinal variable). Second, to identify critical age thresholds where significant shifts in density patterns occur, we implemented a changepoint detection algorithm using the binary segmentation method. Third, we applied k-means clustering to identify natural groupings in the age-density relationship. Fourth, we conducted stratified analyses to examine density patterns in specific age subgroups and to identify outliers (e.g., older women with high density, younger women with low density).

We developed a risk stratification model incorporating age-specific breast density patterns through several sequential steps. We began by creating composite variables through feature engineering to capture the nuanced aspects of the age-density relationship. Multiple candidate models were then developed including ordinal logistic regression, random forest, gradient boosting, and neural networks. To assess model performance, we implemented k-fold cross-validation (k = 10) using standard metrics including Area Under the Curve (AUC).

Based on the risk stratification model, we developed personalized screening protocols through a systematic approach. We first established clinically relevant risk thresholds based on model outputs, creating three risk categories (low, intermediate, high). For each risk category, we then defined recommended screening intervals and modalities based on published guidelines and expert consensus. Finally, we developed a Monte Carlo simulation model to estimate the impact of personalized protocols compared to standard age-based screening. The simulation incorporated parameters from published literature on cancer detection rates, false positive rates, and resource utilization (Loewen et al., 2022).

All statistical analyses were performed using R version 4.2.0 (R Foundation for Statistical Computing, Vienna, Austria) and Python 3.9 with scikit-learn 1.0.2. Simulation modeling was conducted using SimPy 4.0.1.

3. Results

3.1. Demographic and breast density characteristics

Among the 2584 women included in the analysis, the mean age was 53.98 years and Standard deviation (SD = 8.84), with the majority of participants falling within the 40–49 (39.4 %) and 50–59 (35.1 %) age groups. The overall distribution of breast density categories was: category A (predominantly fatty) 22.8 %, category B (scattered fibroglandular densities) 52.8 %, category C (heterogeneously dense) 20.4 %, and category D (extremely dense) 3.9 %. (Lehman et al., 1999)

3.2. Age-density relationship

A significant association was observed between age group and breast density category (p < 0.01). As shown in Table 1, the proportion of women with dense breasts (categories C and D) decreased progressively with increasing age, while the proportion with predominantly fatty breasts (category A) increased. In the youngest age group (<40 years), 60.0 % of women had dense breasts (categories C and D combined), compared to only 5.2 % in the oldest age group (70+ years).

Table 1.

Distribution of Demographic Characteristics and Breast Density Categories Among Women Participating in Breast Cancer Screening (N = 2584), Saudi Arabia, October 2023–December 2024.

Age Group	N (%)	Category A N (%)	Category B N (%)	Category C N (%)	Category D N (%)
<40	35 (1.4)	1 (2.9)	13 (37.1)	16 (45.7)	5 (14.3)
40–49	1019 (39.4)	127 (12.5)	491 (48.2)	332 (32.6)	69 (6.8)
50–59	908 (35.1)	217 (23.9)	519 (57.2)	149 (16.4)	23 (2.5)
60–69	525 (20.3)	192 (36.6)	302 (57.5)	27 (5.1)	4 (0.8)
≥70	96 (3.7)	51 (53.1)	40 (41.7)	4 (4.2)	1 (1.0)
Total	2584 (100)	588 (22.8)	1365 (52.8)	528 (20.4)	102 (3.9)

Category A = Almost entirely fatty; Category B = Scattered fibroglandular; Category C = Heterogeneously dense; Category D = Extremely dense. Note: Association between age group and breast density category was statistically significant (p < 0.01).

Polynomial regression analysis revealed a significant non-linear relationship between age and breast density. The third-degree polynomial model provided significantly better fit than the linear model (p < 0.01). All coefficients were statistically significant (p < 0.01), indicating that the relationship between age and breast density follows a complex non-linear pattern rather than a simple linear decline (Wen et al., 2016).

Changepoint analysis identified three critical age thresholds where significant shifts in density patterns occurred (Table 2). These thresholds at approximately 42, 52, and 65 years represent ages at which the distribution of breast density categories changed most dramatically, suggesting natural transition points for risk stratification (Kerlikowske et al., 2015).

Table 2.

Critical Age Thresholds for Breast Density Transitions Identified by Changepoint Analysis Among Women Screened in Saudi Arabia, October 2023–December 2024.

Threshold	Age (years)	95 % CI	Interpretation
1	42.3	41.1–43.5	Early 40s transition
2	51.7	50.4–53.0	Early 50s transition
3	65.2	63.8–66.6	Mid-60s transition

Note: Thresholds represent ages at which the distribution of breast density categories changed most dramatically, suggesting natural transition points for risk stratification.

K-means clustering identified four distinct clusters in the age-density relationship. Cluster 1 (n = 612) comprised younger women with a mean age of 45.3 years who predominantly had dense breasts (categories C and D). Cluster 2 (n = 953) consisted of middle-aged women with a mean age of 52.7 years who had scattered fibroglandular densities (category B). Cluster 3 (n = 734) included older middle-aged women with a mean age of 58.4 years who had scattered fibroglandular to fatty breasts (categories A and B). Cluster 4 (n = 284) comprised elderly women with a mean age of 67.2 years who predominantly had fatty breasts (category A). Bootstrap validation confirmed the stability of these clusters, with an average silhouette width of 0.68, indicating good cluster separation. Bootstrap validation confirmed the stability of these clusters, with an average silhouette width of 0.68, indicating good cluster separation.

3.3. Risk stratification model

The performance of the four modeling approaches for risk stratification is compared in Table 3. The gradient boosting machine demonstrated the best overall performance with AUC of 0.83 (95 % CI, 0.80–0.86) in the validation set, followed by the neural network, random forest, and ordinal logistic regression.

Table 3.

Performance Comparison of Four Modeling Approaches for Risk Stratification Based on Screening Data from Saudi Arabia, October 2023–December 2024.

Model	AUC (95 %, CI)	Sensitivity %	Specificity %	Accuracy %
Logistic Regression	0.76, (0.73–0.79)	72.3	71.8	72.0
Random Forest	0.80, (0.77–0.83)	75.8	74.6	75.1
Gradient Boosting	0.83, (0.80–0.86)	79.2	78.5	78.8
Neural Network	0.81, (0.78–0.84)	77.5	76.2	76.8

The gradient boosting model showed excellent calibration (Hosmer-Lemeshow test p = 0.42), indicating good agreement between predicted and observed risk. Feature importance analysis revealed that age, breast density category, and their interaction terms were the most influential predictors in the model.

Based on the gradient boosting model, participants were classified into three risk categories (low, intermediate, and high), with the distribution shown in Table 4. The distribution of risk categories varied significantly by age group (p < 0.01). Notably, 42.9 % of women under 40 years were classified as high risk, compared to only 5.2 % of women aged 70 years or older.

Table 4.

Distribution of Risk Categories by Age Group Among Screened Women (N = 2584), Saudi Arabia, October 2023–December 2024.

Age Group	Low Risk n (%)	Intermediate Risk n (%)	High Risk n (%)	Total n (%)
<40	4 (11.4)	16 (45.7)	15 (42.9)	35 (100)
40–49	412 (40.4)	437 (42.9)	170 (16.7)	1019 (100)
50–59	448 (49.3)	352 (38.8)	108 (11.9)	908 (100)
60–69	329 (62.7)	147 (28)	49 (9.3)	525 (100)
≥70	70 (72.9)	21 (21.9)	5 (5.2)	96 (100)
Overall	1248 (48.3)	969 (37.5)	367 (14.2)	2584 (100)

Note: The Chi-squared test was used to test the association. Association between age group and risk category distribution was statistically significant (p < 0.01).

3.4. Personalized screening protocol

Personalized screening protocols were developed for each risk category, specifying screening intervals, recommended imaging modalities, and age-specific considerations (Table 5). Women classified as low risk are recommended to undergo biennial mammography beginning at age 50 and continuing through age 74. Those in the intermediate risk category should receive annual mammography starting at age 45 and continuing to age 74, with supplemental ultrasound considered for women with dense breasts. For high-risk women, the protocol recommends annual mammography with digital breast tomosynthesis beginning at age 40 and continuing through age 74, with supplemental Magnetic resonance imaging considered for those with extremely dense breasts or additional risk factors.

Table 5.

Personalized Breast Cancer Screening Protocols by Risk Category Based on Analysis of Data from Saudi Arabia, October 2023–December 2024.

Risk Category	Age Range	Screening Interval	Modality	Starting Age	Additional Recommendations
Low	All ages	Biennial	Digital Mammography	50	None
Intermediate	All ages	Annual	Digital Mammography	45	Clinical breast exam annually
High	All ages	Annual	Digital Mammography + Ultrasound	40	Magnetic Resonance Imaging (MRI) consideration for highest risk

Monte Carlo simulation comparing the personalized protocols to standard biennial screening for all women aged 50–74 years predicted the following outcomes per 1000 screened women over a 10-year period. Cancer detection increased from 31.2 cases with standard screening to 35.8 cases with personalized protocols, representing a relative increase of 14.7 %. False positives decreased from 412 with standard screening to 372 with personalized protocols, representing a relative reduction of 9.7 %. Fig. 1 illustrates these relative improvements, with personalized screening showing 114.7 % of standard screening's cancer detection rate and 90.3 % of its false positive rate. Sensitivity analyses demonstrated that these results were robust to variations in model parameters within plausible ranges.

Fig. 1.

Comparison of Key Outcome Measures for Standard Versus Personalized Breast Cancer Screening Protocols Based on Monte Carlo Simulation, Saudi Arabia, October 2023–December 2024 Values shown are relative to standard biennial screening (baseline = 100 %). Cancer detection: 114.7 % represents a 14.7 % increase (35.8 vs. 31.2 cases per 1000 women). False positives: 90.3 % represents a 9.7 % decrease (372 vs. 412 per 1000 women). Relative values calculated as (personalized/standard) × 100.

4. Discussion

This study demonstrates a complex non-linear relationship between age and breast density, with significant transitions at specific age thresholds. These patterns have critical implications for developing personalized screening protocols that could improve cancer detection while optimizing healthcare resources.

Our results confirm the inverse relationship between age and breast density but extend beyond this trend by characterizing its specific nature. The third-degree polynomial model as the best fit indicates the decline is not uniform, but follows a complex pattern with varying rates of change (Ming et al., 2019). The three critical age thresholds identified around the early 40s, early 50s, and mid-60s align with key physiological transitions, corresponding to the perimenopausal period, the average age of menopause, and a later-life stabilization of breast density patterns, respectively (Ho et al., 2025). These findings support the need for age-specific approaches to breast density interpretation rather than applying uniform categories across all ages (Lokate et al., 2013).

The superior performance of our gradient boosting model (AUC = 0.83) over traditional logistic regression (AUC = 0.76) highlights the value of machine learning in capturing complex, non-linear risk factor relationships, consistent with recent advances in deep learning for breast cancer risk prediction (Yala et al., 2021). The risk category distribution further challenges conventional age-based screening; 42.9 % of women under 40 were classified as high risk, compared to only 5.2 % of women over 70, underscoring the limitations of using age as a primary screening criterion. This aligns with the conclusions of Walker et al. (2024), who argued that the age-based approach to screening overlooks population heterogeneity in breast cancer risk (Walker et al., 2024).

The developed protocols represent a significant shift from conventional approaches. Our simulation results suggest this strategy could increase cancer detection by 14.7 % while reducing false positives by 9.7 % compared to standard age-based screening (Walker et al., 2025). These findings are consistent with modeling studies by Trentham-Dietz et al. (2016) (Trentham-Dietz et al., 2016), who projected that risk-based screening could improve the benefit-to-harm ratio of mammography. These relative improvements are particularly noteworthy because they occur simultaneously personalized screening achieves both better detection and fewer false alarms. The 14.7 % relative increase in cancer detection translates to approximately 4.6 additional cancers detected per 1000 women screened over 10 years, while the 9.7 % relative reduction in false positives represents 40 fewer women experiencing unnecessary anxiety and additional testing. This dual benefit addresses two of the most significant challenges in breast cancer screening programs.

Translating our findings into clinical practice would require careful consideration of implementation challenges. Implementing these findings requires user-friendly clinical decision support tools and clear patient communication strategies, especially when recommendations diverge from standard guidelines (Azam et al., 2020). The three-category risk stratification system (low, intermediate, high) strikes a balance between precision and practicality. While more granular risk categories might theoretically allow for more personalized recommendations, they could also introduce unnecessary complexity into clinical decision-making. This approach aligns with recommendations from the European Conference on Risk-Stratified Prevention and Early Detection of Breast Cancer (Pashayan et al., 2020), which advocated for simplified risk categories to facilitate implementation (Pashayan et al., 2018).

Several limitations of this study warrant consideration. First, the cross-sectional design prevents direct assessment of how breast density changes over time within individuals. Longitudinal studies would be valuable to validate the age-specific patterns identified in our cross-sectional analysis. Second, our dataset lacks outcome data on cancer diagnoses, which limits our ability to directly assess the impact of personalized screening on cancer detection rates. The simulation results, while informative, rely on assumptions from published literature rather than direct observations (Rainey et al., 2019; Esserman, 2017). Third, our study lacks direct measurement of actual cancer detection and false-positive rates from clinical implementation, which would provide more definitive evidence of the model's real-world effectiveness and enhance understanding of its clinical impact. Lastly, the small sample size in certain subgroups, particularly women under 40 (n = 35), limits the precision of our estimates for these populations.

5. Conclusion

This study provides valuable insights for the value of incorporating age-specific breast density patterns into personalized breast cancer screening approaches. Our analysis of 2584 women revealed a complex non-linear relationship between age and breast density, with three critical age thresholds around the early 40s, early 50s, and mid-60s representing significant transitions in density patterns. The risk stratification model developed using gradient boosting techniques demonstrated excellent performance (AUC = 0.83) in categorizing women into risk groups that varied substantially across age categories. Personalized screening protocols from this model could improve outcomes by increasing cancer detection by 14.7 % and reducing false positives by 9.7 %.

Our findings challenge the one-size-fits-all screening approach, supporting a shift to personalized strategies that use complex, age-specific patterns to better account for individual risk. Implementation of these findings may enhance early detection while optimizing healthcare resource allocation, ultimately contributing to more effective breast cancer control strategies.

Future research should focus on prospective validation of the personalized screening protocols developed in this study. Implementation studies would be particularly valuable to assess the feasibility, acceptability, and effectiveness of these protocols in real-world clinical settings (Tan et al., 2025).

Author contribution

This article is authored by a single individual, [SA]. All aspects of the study, including the conception and design of the research, the literature review, data collection and analysis, and the writing and revision of the manuscript, were performed solely by [SA].

Generative AI and AI-assisted technologies in the writing process

During the preparation of this work the author used ChatGPT & Grammarly in order to proofread this manuscript. After using this tool/service, the author reviewed and edited the content as needed and takes full responsibility for the content of the publication.

CRediT authorship contribution statement

Sahal Alotaibi: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Software, Resources, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization.

Institutional review board statement

The study protocol was approved by the ethical committee at the Taif health cluster ethics committee, approval number: [2025-E-40] on 14/04/2025.

Funding

The author would like to acknowledge the Deanship of Graduate Studies and Scientific Research, Taif University for funding this work.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

Supplementary material

Data availability

Data will be made available on request.