Nomogram predicting the risk of three-year chronic kidney disease adverse outcomes among East Asian patients with CKD

Chronic kidney disease (CKD) is becoming a common global health challenge. A report in 2016 pointed out that the prevalence of all stages of CKD varies between 7 and 12% in the different regions of the world [1]. In 2019, a 15% overall prevalence of CKD was reported in adults in the United States [2]. Patients with CKD have an increased risk of deteriorating into end-stage renal disease (ESRD), requiring dialysis to survive or suffering from complications such as cardiovascular events or even death. With CKD progression and a high risk of other adverse outcomes, the costs for healthcare and therapies are increasing [3]. Accurate risk prediction for CKD adverse outcomes could contribute to individualized decision making, early patient therapy performance, complication reduction and dialysis preparation [4].

There are some risk models predicting the progression and some adverse outcomes of CKD [5]. Many different risk factors were applied in previous models, and the area under the curve (AUC) of these models ranged from 0.56 to 0.94, with calibration from modest to good [6]. However, there were some limitations for the models mentioned above. First, few of these models have been externally validated. Risk model validation includes internal validation (validating in the same population as the model was developed) and external validation (validating in another population). When the AUC performance was generally acceptable-to-good on the derivation population as internal validation, the AUCs were usually modest-to-acceptable in a new population, even though some AUCs for external validation were not reported in most studies. Second, most of the models focus on only one kind of CKD [7, 8], with many risk factors or even some novel professional biomarkers [9], which are complicated and difficult to widely use, especially in grassroots hospitals or community hospitals. Third, the previous models were mainly derived from white populations, but seldom paid attention to East Asian patients with CKD. It was pointed out that the progression and some adverse outcomes of CKD also varied within countries by ethnicity [10]. Black and Asian people in the UK, Hispanics in the USA, and indigenous people in some other developed countries are at higher risk of CKD progression [11]. Given these inconvenience, it is necessary to validate a simple, intuitive and easily applied predictive model to predict adverse outcomes of CKD for the Mongoloid CKD population in East Asia.

Nomograms are a kind of predictive tools, representing statistical predictive model by a number of scales, and generate the probability of a clinical event at appropriate values [12]. Using data from two different populations, we aimed to develop a simple but intuitive model to predict CKD progression for the East Asian CKD population that can be easily and widely implemented for clinical practitioners. To make our findings more credible, the public data were randomly split at a ratio of 7:3, into a developing and another internal validation data set. The enrolled data from our hospital were analysed for external validation. Model validation including discrimination, calibration and clinical values, was assessed in all three datasets.

The characteristics of patients at baseline in the development and validation data sets are listed in Table 1. The development and internal validation data set included 797 patients (191 with adverse outcomes [23.96%]) and 341 patients (89 with adverse outcomes [26.10%]) from CKD-ROUTE data set. The P value of sex, body mass index(BMI), aetiology of CKD, blood pressure(mmHg), serum albumin(g/dL), haemoglobin(g/dL), eGFR (ml/min/1.73 m²), case history, dipstick proteinuria, urinary occult blood, medication usage and adverse outcomes were all > 0.05, showing no significant difference between these two groups. On the other hand, 297 patients (108 with progression [36.36%]) from the First Hospital of Foshan were included in the external validation data set. The variables between development data set and external validation data-set were significantly different with P < 0.05, demonstrating that these two datasets completely varied.

A total of 12 predictive variables before variable selection should ensured that at least 120 events occur. In our development data set, a total of 797 patients and CKD adverse outcomes occurred in 191(23.96%) patients, showing a sample size larger than 120 events, which means that the sample of our predictive model was adequate.

The variables determined from the Cox proportional hazards model for the nomogram predictive model included age, eGFR, haemoglobin, albumin, and dipstick proteinuria after selection by univariable and multivariable Cox proportional hazards models, with P < 0.05(Table 2). The graphical nomogram predicting chronic kidney disease adverse outcomes in 3 years is shown in Fig. 1. The line named three-year adverse-outcome-free probability meant that the patients did not experience adverse outcomes, including > 50% eGFR loss, initiation of dialysis in ESRD, cardiovascular events (CVEs) and all-cause death in 3 years. The C-statistics of this nomogram were 0.90 (95% CI, 0.89–0.92) for the development data set, 0.91 (95% CI, 0.89–0.94) for the internal validation data-set and 0.83 (95% CI, 0.78–0.88) for the external validity data set, all showing good discrimination for our model.

The calibration plots for the development, internal validation and external validation datasets are shown in Fig. A, B and C in Fig. 2. They demonstrated good agreement between the predicted probability and the observed outcome, with good fitness to the ideal standard line (light grey line), revealing good calibration of our simple predictive nomogram model.

The decision curve analysis (DCA) (Fig. 3) curve of our predictive model was farther from the x or y axis than the eGFR curve or other variables alone in our model, which showed better clinical value and potential net benefit in our predictive model.

In our study, the factors associated with CKD adverse outcomes included age, eGFR, haemoglobin, albumin, and dipstick proteinuria. Then, we developed a visualized predictive nomogram model with these five factors to predict 3-year CKD adverse outcomes for the East Asian CKD population. The model validation was good when tested in different aspects and populations, which means this nomogram model was fit for predicting 3-year CKD adverse outcomes and can be easily and widely implemented for clinical practitioners.

In our study, age was found to be a significant risk factor for CKD adverse outcomes, which was different from some previous Asian studies, such as the 2013 [23] and 2019 [24] Taiwan studies. These differences might be due to declining eGFR resulting from loss of muscle mass, reduced physical activity and decreased food intake in elderly patients [25], resulting in age serving as a protective factor in Taiwanese studies. On the other hand, a more common prevalence of CKD was reported in people aged 65 years or older than in people of younger [2]. Furthermore, other CKD risk factors, such as heart disease, obesity, CKD history, and past damage to the kidneys were all more common in the elder.

Proteinuria was also a risk factor for CKD adverse outcomes. In a previous study, proteinuria, a sign of kidney injury, was perceived as the most powerful predictor [26] of kidney damage for promoting inflammation and fibrosis of kidneys [27], which is closely related to a high risk of CKD adverse outcomes. In the MRFIT study, dipstick proteinuria with protein excretion was associated with a greater risk of ESRD [28]. On the other hand, proteinuria was also deemed a strong independent predictor of cardiovascular risk [29], which is the leading cause of death in patients with CKD.

Here, we developed a simple visualized predictive nomogram model to predict CKD 3-year adverse outcomes for East Asian patients and passed internal validation in Japanese patients and external validation in Chinese patients. A 13-year study in Singapore [30] developed a predictive model for CKD in diabetes mellitus, focusing on long-term CKD progression in diabetic nephropathy, which is different from ours. Usually, eGFR is a simple method of monitoring changes in kidney function, estimating CKD adverse outcomes and determining the beginning of dialysis. However, our nomogram showed better discrimination and clinical value than eGFR alone in terms of 3-year adverse-outcome-free probability. In our nomogram, different values of the variables pointed to various corresponding points in line One. For example, consider a 40-year-old patient with an eGFR of 19, haemoglobin 13 g/dl, albumin 3.0 g/dL and 1 dipstick proteinuria; the 40-year age points to approximately 5 points in line 1, while the other factors point to approximately 20 points in line 1, for a total points of 84 points in line 5, which points to a 35% 3-year adverse-outcome-free probability in the last line in Fig. 1. These probabilities could provide physicians and patients with a general scope of prognosis [8] and guide the next clinical steps, such as establishing vascular access for dialysis preparation. Furthermore, a predictive prognosis could offer mental preparation to patients with CKD and help them understand and accept their diseases.

However, our analysis had a few limitations. First, the external validation of our nomogram model was conducted in a retrospective population, but the other data sets were both from a cohort study. Second, the CKD-ROUTE study also reported that factors have certain associations with renal prognosis, but we did not explore them in our model because we did not obtain those data. Further studies are needed. Third, we did not eliminate the possible effect of underlying diseases and medications used for this model. Further studies will be performed and need to be investigated in the future. Fourth, we did not separate the risk of > 50% eGFR loss, initiation of dialysis and other causes of death, but knowing the probability of CKD adverse outcomes is also of benefit to patients and clinical practitioners.

In conclusion, we developed a visualized predictive nomogram model to predict CKD three-year adverse outcomes for East Asian patients with CKD. The model validation was good when tested in different aspects and populations, offering an easy and widely applicable model for clinical practitioners. However, further prospective population-based studies are needed to investigate the mechanisms.