Validation of the ‘Pathobiological Determinants of Atherosclerosis in Youth Risk Score’ in a Norwegian cohort of young adults

Atherosclerosis, the most common pathophysiological cause of cardiovascular (CV) disease, starts early in life and progresses silently through adulthood.1 Whereas validated tools for accurate risk stratification may facilitate early identification of individuals at high CV risk, most established risk equations are only validated for individuals above 40 years of age. The Pathobiological Determinants of Atherosclerosis in Youth (PDAY) risk score was developed by linking CV risk factors assessed post-mortem to the amount of atherosclerotic lesions in 15- to 34-year-old individuals who died from trauma. The PDAY risk score assigns positive points to increasing age, smoking, obesity in men, hypertension, increasing non-HDL cholesterol, and hyperglycaemia, while awarding negative points to protective factors like female sex and high HDL cholesterol, and has later been shown to be associated with future risk of CV events in the Coronary Artery Risk Development in Young Adults (CARDIA) cohort.2,3 Since CV risk varies over time and between populations, our aim was to evaluate the association between the PDAY risk score and risk of future CV events in a Norwegian contemporary cohort of young adults. Although the PDAY risk score accounts for sex differences through negative points for female sex, we report our results stratified by sex to visualize differences in CV risk during follow-up.3,4 The Oslo Health Study (HUBRO) was conducted in Oslo, Norway, from May 2000 to September 2001. As reported previously, all citizens born in 1969 or 1970 were invited, and 6300 (28%) participated.5 For this study, participants with missing personal identification number (n = 334), self-reported CV event prior to study inclusion (n = 16), or missing measurements required to calculate PDAY risk score (n = 88) were excluded. The PDAY risk score was calculated at baseline.2 Similar to the CARDIA study, points for age were excluded, since all participants were in the same age category (30–31 years).3 The participants were followed for the occurrence of CV events until the end of 2024, censoring non-cardiovascular death. The primary outcome was a composite of CV death, non-fatal myocardial infarction (MI), non-fatal ischaemic stroke, and coronary revascularization for causes other than MI.3 CV events were collected from the Norwegian Patient Registry and the Norwegian Cause of Death Registry. Ethical approval from the Regional Committee of Ethics was obtained (REK Sør-Øst, #265625), and participants provided written informed consent before participation. Analyses were performed using STATA 18. Unadjusted hazard ratios (HRs) were calculated using Cox regression for the association between PDAY risk score and CV events. Since the PDAY risk score already includes common CV risk factors, no additional covariate adjustments were performed. Cumulative incidence curves were drawn for the total population, both stratified by sex and by risk score level. The linearity of the association between the PDAY risk score and CV risk was tested with a likelihood ratio test (P = 0.336 for non-linearity). The population was divided into groups with the intention to make three numerically equal groups, acknowledging the discrete categorization of the PDAY score (level 1: −2–0, n = 2266; level 2: 1–4, n = 1957; level 3: 5–24, n = 1639; Figure 1A). HRs for levels 2 and 3 were calculated using level 1 as reference. Testing of proportional-hazard assumptions was based on Schoenfeld residuals and log–log plots, and there were no violations. (A) Histogram of PDAY risk score distribution. Histogram presenting the distribution of PDAY score in the total population, with visualized cut-offs for the three risk score levels. (B, C) Nelson–Aalen cumulative hazard estimates. Cumulative incidence of cardiovascular events by sex as well as by levels of PDAY risk score, with 95% confidence intervals. (D) Calibration plot showing the relationship between predicted and observed risk by deciles of predicted risk. PDAY risk score overestimates cardiovascular risk in the lowest and highest deciles of risk. L1, level 1; L2, level 2; L3, level 3; PDAY, Pathobiological Determinants of Atherosclerosis in Youth. Model discrimination was evaluated with area under the receiver operating characteristic (ROC) curve (AUC), where values above 0.7 were considered acceptable and between 0.8 and 0.9 excellent.6 Model calibration was evaluated using the Hosmer–Lemeshow test and calibration plots. The final cohort in the present analysis included 5862 individuals (55.5% women) aged 30–31 years (mean 30.6 years). The distribution of the PDAY risk score was skewed towards low risk estimates (Figure 1A). The median PDAY risk score was 2 (IQR −1 to 5) in the total population (age excepted), with a higher score among men (median 4, IQR 2 to 8) than women (median 0, IQR −1 to 2) (P-value <0.001). In total, 130 CV events occurred during follow-up. The crude HR for CV events per 1 point increase in PDAY risk score was 1.2 [95% confidence interval (CI) 1.1–1.2] in the total population, 1.1 (95% CI 1.1–1.1) in men, and 1.2 (95% CI 1.1–1.3) in women (Figure 1B). When divided into PDAY risk score levels, the crude HR for CV events during follow-up for levels 2 and 3 compared with level 1 was 3.1 (95% CI 1.7–5.5) and 6.4 (95% CI 3.7–11), respectively (Figure 1C). The unadjusted AUC for PDAY risk score was 0.712 (95% CI 0.668–0.755) in the total population, 0.635 (95% CI 0.581–0.689) among men, and 0.648 (95% CI 0.546–0.749) among women. The PDAY risk score showed good calibration (χ2 = 8.23, P = 0.411 for lack of fit) (Figure 1D). In summary, the PDAY risk score, calculated in a Norwegian cohort of young adults, was associated with future CV events during 24 years of follow-up. The risk estimates were similar in women and men. Discrimination was in the lower end of what is defined as ‘acceptable’, with an AUC of 0.712, not different between the sexes.6 Calibration plot showed a slight overestimation of CV risk in the upper decile (Figure 1D), in contrast to the CARDIA study.3 In comparison, model discrimination of the PDAY risk score in the CARDIA cohort was 0.775.3 This difference may reflect temporary changes in CV risk factor burden and mortality rates, differences in ethnicity and risk factor distribution, as well as different homogeneity between the cohorts.3,7 However, the AUC was comparable with current CV risk prediction models for adults >40 years. For example, SCORE2 and NORRISK-2 had C-statistics of 0.664–0.813 and 0.76–0.82 in their respective validation cohorts.8,9 Importantly, neither of these scores are validated for adults below the age of 40. Strengths of our study are the large, homogenous, well-characterized, and contemporary cohort of young adults, high female attendance rate, long follow-up period, and events retrieved from high-quality Norwegian health registries. Limitations include a relatively low number of incident clinical events, particularly among women, resulting in wide CIs and limited statistical power for sex interaction testing. Hence, possible sex differences cannot formally be rejected. Further, the PDAY risk score does not consider that some risk factors, like diabetes and obesity, may have a relatively greater impact on CV risk in women than in men, yet assigning points for obesity only to men.4 In addition, the PDAY risk score does not include female-specific risk factors.4 Despite these limitations, our findings suggest that the PDAY risk score might be useful for identification of young individuals at high CV risk, facilitating initiation of early preventive strategies. We want to thank all participants of the ‘Oslo Health Study’. Benedikte Lutterbey (Conceptualization, Funding acquisition, Methodology [supporting], Data curation, Investigation [equal], Formal analysis, Writing—original draft, Writing—review & editing [lead]), Eirik Aaseth (Conceptualization, Investigation, Methodology, Writing—review & editing [equal], Data curation [lead], Formal analysis [supporting]), Sigrun Halvorsen (Project administration, Supervision [supporting], Writing—review & editing [equal]), and Jørgen Gravning (Conceptualization, Funding acquisition, Project administration, Supervision [lead], Methodology, Writing—review & editing [equal]) Author B.L. received a research grant from The Research Council of Norway through the Medical Student Research Program at the University of Oslo (grant number 271555/F20). As the Norwegian Data Protection Agency does not allow open access to our data, our data is not publicly available. Upon reasonable request, additional data can be made available through the corresponding author, pending approval from the Norwegian government.