This study developed and validated a new cardiovascular disease risk score for Japanese people with mild CKD by combining traditional risk factors such as age, sex, and smoking status with specific health checks and prescription data to predict the risk of hospitalization for IHD/CVD and major kidney events. The findings revealed that the developed risk score had a significant discriminatory ability for predicting hospitalization for IHD/CVD and major kidney events (mean c-index for hospitalization for IHD/CVD was 0.75 and for major kidney events was 0.69), with higher risk scores indicating an increased risk of cardiovascular and kidney events.

Previous studies have not adequately evaluated the unique risk profile of patients with mild CKD [23]. For example, the FRS and the Seattle Heart Failure Model were designed primarily for non-CKD populations and thus do not take into account risk factors unique to CKD patients [24, 25]. Therefore, these models could not predict the risk of developing cardiovascular disease in patients prone to kidney dysfunction. The model created for this study actively incorporated key risk assessment indicators in people with CKD. This enabled a more accurate prediction of cardiovascular and kidney events in people with mild CKD. The model’s internal reliability was assessed using a mean c-index of 0.75 for predicting hospitalization for IHD/CVD and 0.69 for predicting major kidney events. This score is comparable or superior to existing models [26, 27]. Furthermore, this model is considered comparable or potentially superior to existing models due to several key factors. First, this model was developed to specifically address the limitations of widely used models like the FRS and the Seattle Heart Failure Model, which were not designed for patients with CKD. By incorporating CKD-specific risk factors, our model offers a more accurate prediction of cardiovascular and kidney events in patients with mild CKD. The study by Weiner et al. aimed to assess the utility of the Framingham equations in predicting incident coronary disease specifically in individuals with CKD [6]. This study focused on a population of individuals aged 45 to 74 years without pre-existing coronary disease, using data pooled from the ARIC and CHS trials. In terms of risk prediction and discriminative ability, Weiner et al. found that the Framingham equations had poor accuracy in predicting cardiac events in CKD patients, with C-statistics of 0.62 and 0.60 for 5- and 10-year events in men, and 0.77 and 0.73 in women, respectively. This moderate discrimination suggested that the Framingham equations generally underpredict events in CKD patients. In contrast, the current study’s model achieved a mean c-index of 0.75 for predicting major adverse cardiovascular events and 0.69 for major kidney events, indicating a valid and slightly better discrimination compared to the Framingham equations, particularly for cardiovascular outcomes in CKD patients. This comparison highlights the improved predictive performance of our CKD-specific model over traditional models like the Framingham risk score.

Additionally, the model was developed using readily available data from routine health checks and administrative databases, making it practical for routine clinical use. This accessibility allows healthcare providers to seamlessly integrate the model into daily practice, even in resource-limited environments. By leveraging easily accessible data, the model enhances the management of CKD patients through early detection and targeted intervention, which can reduce healthcare costs associated with advanced treatments like dialysis and transplantation, as preventive measures are generally more cost-effective than treating advanced stages of the disease [28]. A study involving 439 CKD patients found that those classified as “high” risk by FRS were significantly more likely to experience cardiovascular events [7]. The study also showed that adding biomarkers like albumin, hemoglobin, and eGFR, along with echocardiographic parameters, improved predictive accuracy. In contrast, our study developed a new prediction model for mild CKD patients using readily available data, such as age, sex, BMI, and cholesterol levels. Unlike the FRS study, which enhanced its model with specialized markers, our model intentionally excluded such data to maintain simplicity and broader applicability. Despite this, our model achieved a strong c-index of 0.75 for predicting major adverse cardiovascular events, demonstrating robust performance that is practical for use in diverse clinical settings.

Although many countries have national policies and strategies for noncommunicable diseases, there is often a lack of specific policies focused on education and awareness regarding CKD screening, prevention, and treatment. It is crucial to enhance awareness about preventive measures among the general population, healthcare professionals, and policymakers [29]. Moreover, our model was validated using a large cohort of Japanese patients, which is a significant advantage given that many existing models were developed in non-Asian populations. This population-specific validation ensures that the model’s predictions are more accurate and relevant for Japanese individuals.

Our findings indicate that in a population with mild CKD, where traditional risk factors may not fully capture the subtle risk profile, our models can accurately predict these outcomes. In particular, by incorporating specific health check and prescription data into our models, we were able to increase their practical applicability and relevance to everyday clinical practice. This increase in model accuracy and applicability highlights the importance of a tailored approach to managing patients with CKD. Our findings highlight the potential of these models to help with early intervention strategies, particularly in identifying patients who are at higher risk of adverse outcomes and could benefit from more aggressive management or monitoring. Furthermore, the ability of our models to use routine health data provides a significant advantage. Previous studies frequently developed predictive models using variables that were not commonly used in everyday clinical practice and were difficult to obtain [30, 31]. Thus, this study attempted to use variables that are easily applicable in the real world.

The hospitalization for IHD/CVD and major kidney events risk score models were created in this study for risk assessment in patients with mild CKD. These models, which use specific health screening and prescription data, enable more accurate prediction of cardiovascular and kidney outcomes in patients with CKD. This approach not only simplifies the prediction process but also ensures that it is based on widely available data. Such an approach allows for a more personalized risk assessment for each patient, demonstrating its potential to significantly contribute to the implementation of early intervention and prevention strategies. This innovation in risk stratification has the potential to improve patient care by enabling timely and targeted management strategies for those at higher risk. Early detection enables the implementation of interventions aimed at preventing the progression of cardiovascular disease and kidney failure. Specifically, patients with high-risk scores may benefit from more intensive preventive strategies. These include making significant lifestyle changes, carefully managing blood pressure and glucose levels, and starting appropriate pharmacological treatments. For example, the dynamic nature of these models allows for continuous adjustment of patient management plans based on the changing risk profile revealed by regular health screenings and prescription data monitoring. This approach has the potential to not only improve patient outcomes but also increase the precision and responsiveness of CKD management strategies. While age plays a significant role in the predictive power of this model, the value of the model extends beyond age-related predictions for several reasons. It identifies high-risk individuals early, particularly in the elderly, allowing for timely and more effective interventions that can reduce the progression of cardiovascular and kidney disease and improve long-term outcomes. The model’s comprehensive approach is also relevant for younger patients who may be overlooked. By incorporating multiple risk factors such as sex, smoking status, diabetes, hypertension and lipid levels, the model ensures a thorough risk assessment applicable to a broad patient population.

The 5-year risk assessment of hospitalization for IHD/CVD and major kidney events in this study revealed that as scores increased, so did the incidence of both events. In particular, the risk of hospitalization for IHD/CVD increased significantly, from 3% for scores ranging from 0 to 20 to 31% for scores of 36 and higher. Conversely, major kidney events increased from 0% to a maximum of 5% over the same score range, which is significantly lower than the increased risk of hospitalization for IHD/CVD. One of the primary reasons for this difference in incidence between cardiovascular and kidney events is the outcome selection. Cardiovascular events, such as IHD or CVD, are more common and were included as endpoints, while kidney events were specifically defined as reaching kidney failure, not earlier stages like eGFR decline or doubling of creatinine, making them less frequent. This discrepancy can also be attributed to the distinct underlying physiological mechanisms. Cardiovascular events like hospitalization for IHD/CVD are often the result of atherosclerosis and the sudden blockage of blood vessels [32]. These conditions can lead to acute and severe outcomes, particularly in patients with higher risk scores. On the other hand, major kidney events tend to be the result of a gradual and progressive decline in kidney function [33], a process that generally unfolds more slowly and is less common within the timeframe of the study, particularly when the endpoint is strictly defined as kidney failure.

This study has several limitations. First, the study population was primarily Japanese, which may limit the results’ applicability to other ethnic groups. Differences in genetic, environmental, and lifestyle factors can change the risk profiles for CKD and cardiovascular diseases, potentially limiting the model’s global application. Second, the data we used had limitations that may have influenced how broadly our findings apply. We primarily used data from health insurance companies and government healthcare claims. These data may not accurately reflect the variety of personal and lifestyle factors that can influence CKD and its risk. Therefore, we did not account for important factors such as diet, exercise, and socioeconomic status, which could have an impact on the outcomes. Third, the use of qualitative proteinuria for identifying CKD stages 1 and 2 is another limitation. While CKD diagnostic criteria generally recommend quantitative methods, we relied on urine dipstick tests, which may be influenced by urine concentration, potentially leading to variability in diagnosis. Due to the nature of the administrative database used, no additional methods were available to confirm CKD diagnosis. Additionally, the use of ICD-10 codes in claims data introduces potential limitations related to misclassification and missing data. ICD-10 codes, being primarily used for billing, may not always capture the full clinical context, which could result in inaccuracies in identifying CKD stages or related events. Next, while the models developed in this study appear promising, they have not been tested in a real-world clinical setting. Clinical validation through prospective studies or randomized controlled trials is required to confirm the effectiveness of the model in real-world everyday clinical settings. This step is critical for ensuring that healthcare providers can rely on the model to make informed patient care decisions. Finally, the types and dosages of medications prescribed to participants throughout the follow-up period were not fully accounted for. Although certain medications are known to significantly reduce the risk of cardiovascular and kidney events, the available data did not allow these factors to be fully incorporated into the analysis. Consequently, the potential impact of these medications on the observed outcomes may not have been entirely captured. However, given that the study focused on individuals with mild CKD, the overall impact of medications may have been less significant compared to populations with more advanced stages of CKD. Nevertheless, despite these limitations, the study’s strength is its novel approach to developing a cardiovascular disease risk score specifically for Japanese individuals with mild CKD using a comprehensive dataset that includes both health insurance and government healthcare claims. It uniquely incorporates factors such as eGFR and proteinuria, which are important in the early stages of CKD, increasing the model’s predictive accuracy. Furthermore, the use of widely available routine health data makes it suitable for ongoing patient monitoring and management. Therefore, we believe this study is worth publishing. Before these clinical models can be widely implemented, additional external validation is required.