Framingham score adapted: a valid alternative for estimating cardiovascular risk in epidemiological studies

Abstract

Background

Framingham risk score (FRS) is an important cardiovascular risk assessment tool, based on objective measurements of blood pressure and lipid profile, among other factors. However, in large population surveys, these measures are not always available, which limits their use.

Objectives

Evaluate the performance of the FRS predictive results using subjective measures.

Methodology

Cross-sectional study of 1,414 male rotating shift workers in an iron ore extraction company. The original FRS was calculated using objective systolic and diastolic blood pressure measurements, total cholesterol (TC), and HDL cholesterol. The modified FRS was calculated using subjective measurements of blood pressure and lipid profile, based on self-reported medical diagnosis and use of medications for these conditions. Three adaptations were proposed: (1) FRS-SAH, which considers only self-reported hypertension; (2) FRS-DLP, based solely on self-reported dyslipidemia; and (3) FRS-SAH and DLP, which integrates both self-reported factors. Agreement between the two scores was assessed using the kappa coefficient and the Bland-Altman analysis. The accuracy of the scores in predicting cardiovascular risk was compared using the ROC curve and the area under the curve (AUC).

Results

The scatter plot showed a strong correlation (r = 0.9036, p < 0.001) between adapted FRS-SAH and original FRS. The ROC curve showed an AUC with results above 0.85 for all models, confirming the effectiveness of the adapted scale. Bland-Altman indicated good precision between the measurements. Binary logistic regression analysis showed that all the factors associated with CVD-risk by the original FRS were similar to those associated with the adapted FRS. Among the adaptations, the FRS-SAH demonstrated the highest correlation and predictive accuracy.

Conclusion

The adapted FRS proved to be effective in estimating CVD-risk, showing high correlation, sensitivity, specificity, and accuracy compared to the original FRS. Adaptive FRS based on self-reported hypertension, showed the best performance, making it a reliable alternative for contexts where direct measurements are not feasible.

Introduction

Cardiovascular diseases (CVDs) are the leading cause of death worldwide, accounting for estimated mortality rates by region ranging from 73.6 per 100,000 in High-income Asia Pacific to 432.3 per 100,000 in Eastern Europe in 2022 [1]. The prevention and treatment of CVDs depend on the accurate estimation of cardiovascular risk, which is the probability of a person developing a CVD event, such as myocardial infarction, stroke, or heart failure, in a given period [2, 3].

Robust evidence supports the effectiveness of different methods of calculating and classifying cardiovascular risk for the prevention and treatment of CVDs [2, 4]. These cardiovascular risk prediction models are essential for targeting preventive interventions, and maximizing therapeutic benefits in high-risk individuals [5,6,7]. Among the various systems available, the Framingham Risk Score (FRS) stands out as one of the most widely used in clinical practice for cardiovascular risk stratification [8, 9].

The FRS is based on the Framingham Heart Study, a longitudinal cohort study that started in 1948 and followed more than 5,000 participants from Framingham, Massachusetts, USA, to identify risk factors for cardiovascular diseases. Over the decades, findings from this study have provided the foundation for the development of the FRS. The score has been validated and calibrated for different populations and settings and has been incorporated into several clinical guidelines and recommendations. This score uses clinical and laboratory variables that are strongly associated with cardiovascular events. Among the most relevant factors included in the calculation of the FRS are age, sex, total cholesterol (TC) and high-density lipoprotein cholesterol (HDL-c) levels, systolic blood pressure (SBP), smoking habit, and the presence of diabetes mellitus (DM) [8, 9].

Despite its widespread use, the application of the FRS in large-scale epidemiological studies is often hindered by the unavailability of objective blood pressure and lipid profile measurements, which require laboratory testing or specialized equipment. Hypertension and dyslipidemia, two major determinants of cardiovascular risk, are frequently underreported or unavailable in population-based surveys [10, 11]. To address this limitation, alternative methods—such as self-reported hypertension and dyslipidemia, along with medication use—have been investigated as proxies for direct clinical measurements. While self-reported data may be affected by recall bias, studies suggest that self-reported hypertension and dyslipidemia, particularly when combined with medication use, can provide reasonably accurate estimations when validated against clinical records [12, 13].

Given these challenges, this study aims to evaluate the predictive performance of the FRS when incorporating subjective measures of blood pressure and lipid profile, based on self-reported diagnosis and medication use for systemic arterial hypertension (SAH) and dyslipidemia (DLP). By modifying the FRS with these subjective measures, we seek to determine whether this adapted approach can serve as a valid alternative for cardiovascular risk estimation. Additionally, we aim to assess the agreement and accuracy of this modified FRS compared to the traditional score, providing insights into its applicability in large-scale epidemiological studies and settings where objective measurements may be impractical.

Methodology

Design and participants

The preparation and design of this study adhered to the guidelines set forth by the Strengthening Reporting of Observational Studies in Epidemiology (STROBE) initiative. A population of shift workers from an iron ore extraction company in Minas Gerais and Pará, Brazil, participated in the study. The same data were gathered from shift workers in various locations across the nation as part of a larger study called the “Fatigue Management Project” using three cross-sectional panels, with no overlap of participants between them. Three distinct regions of the country were used to conduct the cross-sectional panels: (a) 337 shift workers from four mines in the Iron Quadrangle region participated in the first panel in 2012; (b) 192 shift workers from another mine in the Iron Quadrangle region participated in the second panel in 2015; and (c) 932 shift workers from the southern Pará region participated in the third panel, which was conducted in 2018. As a result, 1461 shift workers in all were evaluated. Due to insufficient statistical power to accurately represent the female population of rotating shift workers and to eliminate potential sex-specific confounding variables, all 38 women were excluded from the study. We improved our study’s internal validity in this way. Furthermore, nine individuals were excluded because they did not have complete information on the variables assessed in the current study, with a total of 1,414 workers evaluated. The workers’ shifts in hours were contingent upon the state. They put in a weekly shift of four cycles and one day off in Minas Gerais. In Pará, there was a five-cycle work week consisting of two rest days, an eight-hour workday, and twenty-four hours of rest.

Data collection

Data collection adhered to established protocols designed to ensure consistency and reliability across all study phases [10,11,12,13,14,15]. Trained professionals conducted in-person assessments, including anthropometric measurements, biological sample collection, and questionnaire administration. These procedures were standardized and carried out during all three cross-sectional panels.

Sociodemographic data

Participants were grouped into four age categories (20–29, 30–39, and 40 years or older). Self-reported skin color was categorized as white or non-white (including black, brown, yellow, or indigenous). Educational attainment was classified into levels (completed primary school, secondary school, technical school, or university degree). The duration of shift work experience was dichotomized as less than five years or five years and above.

Clinical data

A semi-automatic digital device (Omron Healthcare Co., Kyoto, Japan) validated by international standards was used by professionals with the necessary training to take blood pressure [16]. According to the Brazilian Society of Cardiology, hypertension is defined as SBP ≥ 140 mmHg or diastolic blood pressure (DBP) ≥ 90 mmHg. The values were calculated by averaging three measurements [17]. The following steps were taken to take the participant’s blood pressure: first, the participant was given instructions not to smoke, drink coffee or alcohol, or exercise for at least 30 min before the measurement; second, the participant was seated in a comfortable chair with his feet on the floor and his back supported; third, the participant’s arm was supported on a table at the level of his heart; and, fourth, the participant’s arm circumference was used to determine the appropriate cuff size, which was then wrapped around the arm without interference from clothing; fifth, the participant was given a minimum of five minutes of rest before the first measurement was made; final, two measurements were taken at intervals of one to two minutes, and the average was computed and recorded as the blood pressure value.

Biochemical data

Lipid profile, vitamin D, and glycemic analysis were used to evaluate the biochemical profile. Blood samples were taken in the first two studies (2012 and 2015) following a 10-hour fast. 2018 saw operational problems with the company that prevented the fasting collection from taking place. An experienced practitioner extracted blood by making a puncture in the cubital fossa while the subject was seated and their arm rested on a table. Blood samples were taken in tubes with EDTA and fluoride as anticoagulants for the lipid profile and glucose analysis. The samples were kept at -20 °C until analysis after being centrifuged for 10 min at 3000 rpm. Enzymatic techniques were used to measure the lipid profile (total cholesterol, HDL cholesterol, and triglycerides) and glucose using an automated analyzer (Cobas Integra 400 plus, Roche Diagnostics).

The method used to determine the lipid profile was enzymatic-colorimetric. Friedewald (1972) provided the formula for calculating the low-density lipoprotein fraction (LDL-c), which is LDL-c (mg/dL) = TC - HDL-c - (Triglycerides/5) [18]. When employees’ data included at least one of the following factors—TC > 190 mg/dL, triglycerides 150 > mg/dL (with fasting) or triglycerides > 175 mg/dL (without fasting) or LDL-c > 130 mg/dL or HDL-c < 40 mg/dL— DLP was deemed to exist [17].

Original Framingham risk score

The following factors are included in the original FRS calculation: age, TC, HDL-c, diabetes, smoking, and SBP (both treated and untreated). Given that the progression and incidence of cardiovascular diseases are influenced by biological and epidemiological differences between the sexes, each of these variables is scored by distinct tables for men and women. As a result, a score is given to each risk factor according to the measured values of blood pressure, cholesterol, diabetes, and smoking. We used all of the variables that were suggested and previously described as collected to calculate the initial score. A value that is linked to a percentage that predicts the risk of cardiovascular events over the following ten years is produced by adding up these scores [8, 9].

Adapted Framingham risk score

The adapted FRS maintained the logic of the original score, which considers SBP and lipid profile to be an independent risk factor for CVD. However, in this adapted version, instead of using direct SBP and lipid profile measurements, information on self-reported SAH or DLP and the use of antihypertensive or lower-lipid drugs were used as proxies.

The pre-existing diseases evaluated were SAH, and DLP, by self-report of medical diagnosis and/or use of medications for these conditions using a standardized questionnaire. Medications were classified according to the ATC-Anatomical Therapeutic Chemical, a system of alphanumeric codes developed by the World Health Organization (WHO) for the classification of drugs and other medical products [19]. The drugs evaluated were: antihypertensives, with ATC code C (cardiovascular drugs); lipid-lowering drugs, with ATC code C10 (drugs that act on lipid metabolism); hypoglycemic drugs, with ATC code A10 (drugs used in diabetes); and antiplatelet drugs, with ATC code B01 (antithrombotic drugs).

Participants who reported having SAH or DLP but were not on drug treatment received the highest score, with 4 points for SAH and 5 points for DLP. This reflects the severity of untreated morbidities and takes into account the difference in the impact of CVD risk. On the other hand, individuals who reported having these morbidities and were being treated with medication, or who, even without reporting these morbidities, were using these medications, received an intermediate score of 2 points. Finally, those who did not report these morbidities and were not using lowering drugs received the lowest score, i.e. 0 points for both morbidities (Table 1).

Table 1 Point allocations for Framingham risk score after adaptation

This adaptation of the FRS took into account the protective effect of drug treatment in reducing SBP or DLP and, consequently, in reducing cardiovascular risk. Thus, three adapted scores were developed: Adapted FRS-SAH, based only on self-reported SAH; Adapted FRS-DLP, based only on self-reported DLP; and Adapted FRS-SAH and DLP, based on a combination of self-reported SAH and DLP.

That said, CVD-risk in all FRS scores, was categorized dichotomously, with participants classified as “low risk” (CVD-risk < 5%) or “intermediate to high risk” (CVD-risk ≥ 5%).

Statistical analysis

For statistical analyses, we used the Stata/MP program (version 15.0), assuming an alpha of 5%. To characterize the sample among the outcomes analyzed, Pearson’s chi-square test was performed.

ROC curve analyses were conducted to evaluate the performance of the adapted FRS compared to the original FRS, with the original FRS serving as the reference standard. To quantify the ability of the adapted FRS to discriminate between different levels of cardiovascular risk, two classifications were considered: the first distinguishing between low-risk individuals and those at intermediate to high risk, and the second separating low and intermediate-risk individuals from those at high risk. This dual approach was adopted to create binary variables that facilitate the construction of 2 × 2 contingency tables, enabling a detailed analysis of the predictive performance of the adapted FRS across different risk scenarios. This method allows for a comprehensive evaluation of its accuracy in identifying varying levels of cardiovascular risk.

Concordance between the original FRS and the adapted FRS was further assessed using the Bland-Altman analysis. This method evaluates the agreement between two measurement techniques by identifying any systematic bias and the variation around this bias. The mean difference between the two scores was calculated and plotted in a Bland-Altman graph, which displays these differences relative to the average of the scores. Limits of agreement (LoA) were defined as the mean difference ± 1.96 times the standard deviation of the differences. This analysis helps in evaluating the consistency of differences across the range of measurements, identifying any significant deviations or biases between the original and adapted scores.

The accuracy of the adapted FRS in predicting cardiovascular risk was examined using specific cut-off points. Sensitivity, or the ability to correctly identify true cases, and specificity, or the ability to correctly exclude unaffected cases, were analyzed along with overall accuracy, which represents the proportion of correct diagnoses (both positive and negative). Accuracy was determined using the formula ((TP + TN) / (TP + TN + FP + FN)), where TP denotes true positives, TN denotes true negatives, FP denotes false positives, and FN denotes false negatives.

Finally, crude binary logistic regression models were applied separately for the original and adapted FRS to examine the association between sociodemographic and health factors and cardiovascular risk, classified as ‘low’ versus ‘intermediate to high’. These models were not adjusted for confounders, as the objective was to assess whether the associations observed with the original FRS maintained their effect magnitude in the adapted FRS when analyzed independently. Furthermore, a multivariable logistic regression model was applied using a stepwise backward approach, including variables with p < 0.20 in the univariate analysis and those deemed biologically plausible. The outcome variable, cardiovascular risk, was categorized as low vs. intermediate-to-high based on the original and adapted Framingham Risk Score instruments.

To quantitatively assess the concordance between the odds ratios (ORs) obtained from the two models, we compared their 95% confidence intervals (CIs). If the CIs overlapped, we considered that there was no statistically significant difference between the ORs of the two models. Additionally, the Breslow-Day test for homogeneity of odds ratios was applied to formally evaluate whether the associations observed with the original and adapted FRS were significantly different. A non-significant result in this test indicates that the ORs are statistically homogeneous between the models.

Results

Table 2 presents the sociodemographic, clinical, and behavioral characteristics of the assessed adults, categorized according to cardiovascular risk. The variables include age, skin color, education level, anthropometric data, and self-reported chronic conditions, among others. The data are expressed in percentages and distributed across low, intermediate, and high-risk categories as defined by the FRS.

Table 2 Sociodemographic, clinical, and behavioral characteristics of shift workers according to cardiovascular disease risk

A total of 1,414 workers were assessed in this study, predominantly aged 20–34 (45.1%), brown (62.2%) and with a high school degree (71.4%). Most of the participants had more than 5 years of professional experience (76.0%) and were from the state of Pará (63.3%).

Anthropometric data showed that 44.4% had a WC ≥ 94 cm and 72.3% had a body mass index (BMI) ≥ 25 kg/m². Self-reported chronic conditions included diabetes (2.8%), dyslipidemia (12.7%), and hypertension (9.5%). Behavioral variables revealed that 48.7% had low physical activity, 15.6% were smokers, and 62.5% consumed alcohol.

In the analysis of the dispersion of scores and linear regression between the original FRS and the adapted FRS, a strong and highly significant positive correlation was observed. The highest correlation was found for the adapted FRS-SAH (r = 0.9036; p < 0.001), followed by the adapted FRS-SAH and DLP (r = 0.7487; p < 0.001) and the adapted FRS-DLP (r = 0.7399; p < 0.001) (Fig. 1).

Fig. 1

Dispersion of scores and linear regression between the original scale and the adapted Framingham risk score with self-reported hypertension and lipid profile. The graphs show the Pearson correlation coefficients between the adapted and original Framingham Risk Score (FRS). (A) Adapted FRS-SAH (self-reported hypertension) vs. Original FRS. (B) Adapted FRS-DLP (self-reported lipid profile) vs. Original FRS (C) Adapted FRS-SAH and DLP (self-reported hypertension and lipid profile) vs. Original FRS

Figure 2 presents the ROC curve comparing the adapted FRS with the original FRS in predicting CVD-risk based on self-reported hypertension and lipid profile. The adapted FRS-SAH, which relies only on self-reported hypertension, showed an AUC of 0.9768 for distinguishing low-risk individuals from those at intermediate/high risk, and 0.9927 for differentiating low/intermediate risk from high risk. The adapted FRS-DLP, based solely on self-reported lipid profile, had an AUC of 0.9279 for the low vs. intermediate/high-risk classification and 0.9727 for low/intermediate vs. high risk. When both self-reported parameters were combined (FRS-SAH and DLP), the AUC values were 0.9471 and 0.9634 for the respective risk classifications. These results highlight the strong discriminative power of the adapted models in assessing cardiovascular risk.

Fig. 2

ROC curve of the adapted scales versus the original Framingham risk score. The graphs show the Receiver Operating Characteristic (ROC) curves comparing the Framingham adapted risk score with self-reported health data to the original Framingham risk score. (A) Adapted FRS-SAH: CVD-risk classification– low vs. intermediate/high risk. (B) Adapted FRS-SAH: CVD-risk classification– low/intermediate vs. high risk. (C) Adapted FRS-DLP: CVD-risk classification– low vs. intermediate/high risk. (D) Adapted FRS-DLP: CVD-risk classification– low/intermediate vs. high risk. (E) Adapted FRS-SAH and DLP: CVD-risk classification– low vs. intermediate/high risk. (F) Adapted FRS-SAH and DLP: CVD-risk classification– low/intermediate vs. high risk

Figure 3 evaluates the agreement between the scales using the Bland-Altman method, indicating moderate agreement between the scores, with bias close to zero, with more accurate results for the adapted FRS-SAH scale, and suggesting that there is no significant systematic deviation between the objective and subjective measurements. The limits of agreement defined above and below the line of average agreement showed that most of the differences between the measurements fell within this range, with a few data points exceeding these limits.

Fig. 3

Evaluation of the concordance of Framingham risk scores by Bland-Altman analysis. In the Bland-Altman graph, the differences between the scores are represented on the y-axis, while the average of the scores is plotted on the x-axis. The horizontal line at the level of the bias indicates the average difference, and the additional horizontal lines represent the upper and lower limits of agreement. These limits are calculated as the mean difference plus and minus 1.96 times the standard deviation of the differences, providing a range in which 95% of the differences are expected to fall if the two methods are equivalent. Points outside these limits suggest disagreement between the measurement methods. The adapted Framingham Risk Score (FRS) was calculated using self-reported health data: (A) FRS-SAH: Based only on self-reported hypertension (hypertension diagnosis and medication use). (B) FRS-DLP: Based only on self-reported lipid profile (dyslipidemia diagnosis and medication use). (C) FRS-SAH and DLP: Combined self-reported hypertension and lipid profile

Table 3 compares the performance of the adapted and original Framingham models. The adapted FRS-SAH showed the best performance, with high sensitivity (89.0%), specificity (93.8%), and accuracy (92.5%) for low vs. intermediate/high risk. It also achieved 100.0% sensitivity and 92.4% accuracy for low/intermediate vs. high risk. The FRS-DLP had lower sensitivity (71.3%) but maintained high specificity (92.1%) and accuracy (86.4%). In the second classification, its sensitivity dropped to 60.0%, while specificity remained high (97.3%). The combined FRS-SAH and DLP had the lowest sensitivity (58.5% and 30.0%) but the highest specificity (95.5% and 97.9%), with accuracy around 87%. These results highlight the FRS-SAH as the most effective adaptation, particularly when self-reported hypertension is used, reinforcing its applicability in cardiovascular risk assessment.

Table 3 Predictive values of adapted Framingham risk scores compared to Framingham’s original risk score

Table 4 shows the results of the binary logistic regression analysis between the FRS original and adapted FRS models with sociodemographic and health variables. It was observed that all the factors associated with CVD-risk by the original FRS were similar to those associated with the adapted models, such as skin color, education, working time, mining regions, waist circumference, body mass index, self-reported chronic diseases, physical activity, smoking and alcohol consumption. Differences in odds ratios were observed between the adaptations, particularly for FRS-DLP FRS-SAH, and DLP, which showed distinct associations with variables such as skin color, education, and mining region. In the adjusted analysis (supplementary material), longer working time (> 5 years), elevated waist circumference (≥ 94 cm), self-reported dyslipidemia, and smoking were significantly associated with increased CVD risk across all Framingham Risk Score models. Higher education levels were protective, with individuals holding a technical or second-degree education showing lower odds of intermediate-to-high CVD-risk (p < 0.001).

Table 4 Logistic regression of socioeconomic and health characteristics with cardiovascular risk according to the Framingham risk score instrument used

To quantitatively assess the concordance between the odds ratios (ORs) obtained from the two models, we compared their 95% confidence intervals (CIs). Most characteristics showed overlapping CIs across models, indicating no statistically significant differences in their associations with cardiovascular risk. Differences emerged for certain variables. Brown-skinned individuals had a lower risk in the models adapted for dyslipidemia (DLP) and dyslipidemia with systemic arterial hypertension (SAH and DLP), but not in the original or SAH-adapted FRS. Similarly, alcohol consumption more than once a week was significantly associated with increased cardiovascular risk in the DLP, and SAH and DLP models but not in the others. These findings highlight that while the overall patterns of association remained stable, specific risk factors varied depending on the adaptation of the FRS used.

Discussion

This study evaluated the performance of a modified FRS, which utilizes subjective measures — derived from self-reported hypertension or dyslipidemia diagnoses and antihypertensive or lower-lipid drug use — and compared its predictive accuracy with the original FRS, which relies on objective measurements. The results demonstrated that the adapted FRS showed a strong correlation with the original FRS, confirming the adapted scale’s effectiveness. Among the adaptations, the FRS-SAH demonstrated the highest correlation and predictive accuracy, making it the most reliable alternative when direct measurements are not available. This validation is particularly significant in settings where direct measurement is unfeasible, such as during the COVID-19 pandemic or in remote and large-scale epidemiological studies. The adaptation of the FRS to incorporate self-reported data allows for the continuation of cardiovascular risk assessment, even under constrained conditions, without compromising the primary objectives of public health monitoring.

The FRS has long been established as a robust tool for predicting cardiovascular risk across diverse populations and settings. Its reliability has been consistently validated, including in multiethnic cohorts such as the one studied by Chia et al. (2015), confirming its predictive accuracy even beyond the original Framingham cohort [20]. Studies like those by Aguilar et al. (2006) and Hippisley-Cox et al. (2007) have further reinforced the FRS’s strength, demonstrating its ability to effectively detect cardiovascular risk in hypertensive populations and maintain competitive accuracy when compared to newer models like QRISK [21, 22]. Even when incorporating additional data, such as ambulatory blood pressure, Bell et al. (2014) showed that the FRS continues to provide reliable predictions, underscoring its utility as a stand-alone tool [23].

The use of self-reported health data in CVD risk assessment has been validated in previous studies showing that they can serve as reliable alternatives when objective measurements are unavailable [24,25,26]. Higashiyama et al. (2007) demonstrated that self-reported hypertension has a high specificity and significant predictive power for cardiovascular mortality, even when compared with clinically measured blood pressure [24]. In a high-risk population, Dey et al. (2015) found that self-reported cardiovascular risk factors, including hypertension, had good sensitivity and specificity, particularly for hypertension and diabetes. Although some discrepancies exist, these self-reported measures can be sufficiently accurate for large-scale epidemiological studies [25]. Similarly, Barroso et al. (2018) in a population sample from the region of Girona, Spain, and a similar proposal to our work, using the FRS, showed that self-tracking methods for cardiovascular risk factors, including blood pressure and cholesterol, performed well when compared to professional measurements, making self-reported data a viable option in contexts where clinical measurements are not feasible [26].

Although the results indicate that the modified FRS has significant discriminatory power, care must be taken due to certain limitations. First, the cross-sectional design is a key limitation, as it does not allow for causal inferences or the long-term validation of cardiovascular risk prediction, which is typically achieved through prospective cohort studies. The Framingham Risk Score was originally validated in longitudinal studies where participants were followed for decades. In contrast, our findings provide a snapshot of cardiovascular risk in shift workers, necessitating further studies to evaluate the long-term predictive value of the adapted FRS.

Another important limitation is the reliance on self-reported data for hypertension and medication use. Although self-reporting is a practical solution in epidemiological studies, it introduces recall bias, inaccuracies in self-reporting, and variations in treatment adherence. To mitigate these issues, structured interviewer training was conducted, validated instruments were used, recall cues were incorporated, and cross-verification of medication use was encouraged. Nonetheless, these measures do not eliminate self-reporting biases, which should be considered when interpreting the findings.

The generalizability of our findings is another important consideration. As the study focuses on male rotating shift workers in Brazil, the applicability of the adapted FRS to other populations, such as women, individuals with different occupational settings, or diverse socioeconomic backgrounds, requires further validation. While the findings highlight important cardiovascular risk factors in this occupational group, they may not be directly applicable to other populations with different occupational, socioeconomic, or demographic characteristics. However, our study underscores the need for tailored cardiovascular risk screening in high-risk occupational settings, particularly for shift workers who experience unique physiological stressors. Future research should explore the applicability of the adapted FRS in broader populations and assess its integration into workplace health programs, similar to initiatives in the mining sector [27].

Despite these limitations, the study has several strengths that enhance the validity of its results. The high predictive accuracy of the adapted FRS, particularly the FRS-SAH, reinforces its reliability as an alternative measure in settings where objective assessments are not feasible. The relatively large sample size reinforces statistical power, facilitating more reliable generalizations within the studied context. Additionally, the use of data from multiple cross-sectional panels over several years strengthens the robustness of the analysis, providing a detailed assessment of cardiovascular risk factors and trends in shift workers.

Conclusion

The modified FRS validation using subjective blood pressure and lipid profile measurements is a valuable alternative for assessing cardiovascular risk when direct measurements are not available. Among the adaptations tested, the version based exclusively on self-reported hypertension (FRS-SAH) showed the best performance, with greater correlation, accuracy and discriminative capacity compared to the original FRS. Including the self-reported lipid profile (FRS-DLP and FRS-SAH and DLP) resulted in lower sensitivity, suggesting that using self-reported hypertension alone may be the best approach when laboratory data is not available. Although the modified FRS shows promise, it is essential that researchers are aware of its limitations and supplement subjective data with additional information when possible. This approach not only facilitates the continuous assessment of cardiovascular risk in various contexts, but also underlines the importance of developing future methodologies in epidemiological research.