Visual Abstract
Keywords: end stage kidney disease, risk factors, diabetes mellitus, epidemiology and outcomes, epidemiology and outcomes
Abstract
Background and objectives
Individuals with type 2 diabetes are at a higher risk of developing kidney failure. The objective of this study was to develop and validate a decision support tool for estimating 10-year and lifetime risks of kidney failure in individuals with type 2 diabetes as well as estimating individual treatment effects of preventive medication.
Design, setting, participants, & measurements
The prediction algorithm was developed in 707,077 individuals with prevalent and incident type 2 diabetes from the Swedish National Diabetes Register for 2002–2019. Two Cox proportional regression functions for kidney failure (first occurrence of kidney transplantation, long-term dialysis, or persistent eGFR <15 ml/min per 1.73 m2) and all-cause mortality as respective end points were developed using routinely available predictors. These functions were combined into life tables to calculate the predicted survival without kidney failure while using all-cause mortality as the competing outcome. The model was externally validated in 256,265 individuals with incident type 2 diabetes from the Scottish Care Information Diabetes database between 2004 and 2019.
Results
During a median follow-up of 6.8 years (interquartile range, 3.2–10.6), 8004 (1%) individuals with type 2 diabetes in the Swedish National Diabetes Register cohort developed kidney failure, and 202,078 (29%) died. The model performed well, with c statistics for kidney failure of 0.89 (95% confidence interval, 0.88 to 0.90) for internal validation and 0.74 (95% confidence interval, 0.73 to 0.76) for external validation. Calibration plots showed good agreement in observed versus predicted 10-year risk of kidney failure for both internal and external validation.
Conclusions
This study derived and externally validated a prediction tool for estimating 10-year and lifetime risks of kidney failure as well as life years free of kidney failure gained with preventive treatment in individuals with type 2 diabetes using easily available clinical predictors.
Podcast
This article contains a podcast at https://dts.podtrac.com/redirect.mp3/www.asn-online.org/media/podcast/CJASN/2022_11_04_CJN05020422.mp3.
Introduction
Worldwide, the prevalence of type 2 diabetes is rapidly increasing (1). Individuals with type 2 diabetes have three to five times higher risk of developing kidney failure compared with individuals without type 2 diabetes (2). Treatment options to prevent or delay kidney failure in individuals with type 2 diabetes include smoking cessation (3); intensive glucose and BP lowering (4,5); and treatment with angiotensin-converting enzyme inhibitors or angiotensin II receptor blockers (6), sodium-glucose cotransporter-2 inhibitors (SGLT2is) (7), and glucagon-like peptide-1 receptor agonists (GLP1-RAs) (7). The absolute benefit that an individual may derive in terms of kidney failure risk reduction from these treatments depends on several different factors, including risk factor burden, duration of treatment, and overall life expectancy.
Few prediction models exist for kidney failure in individuals with type 2 diabetes (8–11). These models have important limitations because they predict risk over a relatively short time period and often predict intermediate outcomes, such as doubling of serum creatinine (8), which might be less relatable to an individual than kidney failure as a hard outcome. Notably, shared risk factors associated with kidney failure also contribute to high cardiovascular disease and mortality risks (12). Therefore, it is crucial to take all-cause mortality into account as a competing risk to avoid overestimation of kidney failure risk because most individuals with type 2 diabetes will die from other causes before developing kidney failure. These gaps highlight the need to develop prediction models for long-term risk of kidney failure in individuals with type 2 diabetes.
Therefore, the aim of this study was to develop and validate a prediction model for the risk of kidney failure in large population-based cohorts of individuals with type 2 diabetes. Further, we aimed to predict life expectancy free of kidney failure and include treatment effects of preventive therapy.
Materials and Methods
Data Sources and Participants
The prediction model was developed and internally validated in the Swedish National Diabetes Register (NDR; n=707,077), which includes individuals with both incident and prevalent type 2 diabetes. Participants in NDR were included from January 1, 2002 until September 25, 2019.
The model was externally validated in an extract of the Scottish Care Information–Diabetes (SCI–Diabetes) database (n=256,265), which includes individuals with incident type 2 diabetes. Participants from SCI–Diabetes were included if their date of diagnosis of diabetes was between January 1, 2004 and January 1, 2019. Both registers have close to complete coverage of the population with a diagnosis of type 2 diabetes during the study period. Register details for both cohorts have been described elsewhere (13,14). All participants were aged >30 years at cohort entry and had a diagnosis of type 2 diabetes (Supplemental Table 1) without kidney failure at baseline. All use of data from these registers received appropriate local data governance approvals, and all studies complied with the Declaration of Helsinki.
Predictor and Outcome Variables
Kidney failure was defined as CKD stage 5 (sustained eGFR of <15 ml/min per 1.73 m2), long-term dialysis, or kidney transplantation (15), and all-cause mortality was defined as death from any cause. Linkage of NDR and SCI–Diabetes to national death registrations and hospital admission/discharge registries enabled the identification of kidney failure using ICD-10 and procedure codes (Supplemental Table 2).
Predictors were preselected on the basis of existing risk scores for kidney failure (9–11) and their availability in clinical practice. Preselection of variables was applied to prevent overfitting (16). The predictors included age, sex (men/women), current smoking (yes or no), systolic BP, body mass index, hemoglobin A1c, eGFR (17), non-HDL cholesterol, albuminuria (none, moderate, or severe), duration of type 2 diabetes (years since diagnosis), insulin treatment (yes or no), and history of cardiovascular disease (yes or no) (Supplemental Table 1). Non-HDL cholesterol was chosen as a single marker to represent the lipid profile (18). Albuminuria was defined as a urine albumin-creatinine ratio of 3–30 mg/mmol for moderate albuminuria and urine albumin-creatinine ratio of >30 mg/mmol for severe albuminuria. An individual’s baseline was set as the date of the first eGFR measurement following enrollment in NDR or diagnosis of diabetes in SCI–Diabetes, and the values of other predictors were defined at the first measurement within 12 months after this date.
Statistical Analyses
Baseline characteristics are described as median and interquartile range (IQR) for continuous variables and as count (percentage) for categorical variables.
Development of the Prediction Model.
A split-sample approach was used for the development and internal validation of the prediction model. A random sample of 75% of participants from NDR (n=530,308) was used as the development dataset. Missing data were imputed using single imputation with predictive mean matching. Details are described in Supplemental Material.
In the derivation dataset, two Cox proportional hazards functions with left truncation and right censoring were developed using age as the time axis: one for prediction of kidney failure events (function A) and one for prediction of all-cause mortality (function B).
Baseline hazards for kidney failure (function A) were derived using 1-year intervals (due to the low amount of kidney failure events), and thereafter, they were smoothed and interpolated to 3-month intervals. Baseline hazards for all-cause mortality (function B) were derived using 3-month intervals (Supplemental Figure 1).
By combining the coefficients from the Cox proportional hazards functions A and B and the smoothed baseline hazards, kidney failure–free survival, 10-year and lifetime risks of kidney failure, and all-cause mortality were calculated using previously validated life tables (19). The 10-year risk of kidney failure was calculated by summation of the predicted kidney failure and all-cause mortality risks in the first 10 years and beyond from a person’s age at cohort entry. Similarly, lifetime risk of kidney failure was calculated by the summation of the predicted kidney failure risks from an individual’s age at cohort entry until the maximum age of 95 years. All analyses were performed with R-statistic programming (version 4.0.3; R Foundation for Statistical Computing, Vienna, Austria). A detailed description of statistical methods is provided in Supplemental Material.
Model Validation for 10-Year Predictions.
Goodness of fit was assessed in the remaining 25% of NDR by calibration plots. Observed risks of kidney failure were calculated using cumulative incidence functions, with the competing event being all-cause mortality. For external validation in SCI–Diabetes, the models were recalibrated on the basis of the incidence of kidney failure and all-cause mortality using the expected versus observed ratios. The logarithm of the expected versus observed ratio was subtracted from the linear predictor for both outcomes. Discrimination was quantified using the Harrell c statistic for survival data (20). Our approach to model development and validation complies with PROBAST guidelines (21) and TRIPOD (22).
Prediction of Treatment Effects.
To estimate the individual treatment benefit, the linear predictor for function A was combined with hazard ratios (HRs) from the most recent high-quality meta-analyses describing effect sizes for each intervention. For this study, we derived estimates of the effect of glucose lowering, BP lowering, GLP1-RA, SGLT2i, angiotensin-converting enzyme inhibitor/angiotensin II receptor blocker treatment, and smoking cessation as described in Supplemental Material. The HRs of smoking cessation, BP lowering, and initiation of GLP1-RA or SGLT2i for all-cause mortality were added to the linear predictor for model B (5,23–25).
The lifetime benefit of treatment was calculated as the difference between predicted median kidney failure–free life expectancy with and without treatment. Similarly, 10-year absolute risk reduction was estimated by calculating the difference between the predicted 10-year kidney failure risk with and without treatment. This same approach was used for estimating lifetime kidney failure risk reduction with initiation of treatment. All model assumptions are provided in Supplemental Table 3.
Sensitivity Analyses
To incorporate the natural decline of eGFR in the predictions of kidney failure risk, several sensitivity analyses were performed to incorporate functions of eGFR over time (Supplemental Material).
Results
Baseline Characteristics
Selection of the development and validation cohorts in NDR is illustrated in Supplemental Figure 2. The NDR cohort consisted of 401,433 (57%) men, median (IQR) age was 65 (IQR, 57–74) years, and median (IQR) eGFR was 85 (IQR, 68–97) ml/min per 1.73 m2 (Table 1). In SCI–Diabetes, 145,753 (57%) were men, median (IQR) age was 61 (IQR, 52–70) years, and median (IQR) eGFR was 83 (68–96) ml/min per 1.73 m2. In NDR, median (IQR) follow-up was 6.8 (IQR, 3.2–10.6) years, with 8004 individuals (1%) developing incident kidney failure and 202,078 (29%) deaths. In SCI–Diabetes, median (IQR) follow-up was 5.9 (IQR, 2.6–9.6) years, with 1653 (0.7%) kidney failure events and 45,056 (18%) deaths.
Table 1.
Baseline characteristics of participants identified from the Swedish National Diabetes Register and the Scottish Care Information–Diabetes cohort after imputation of missing data
| Clinical Characteristics | Swedish National Diabetes Register, n=707,077 | Scottish Care Information–Diabetes Cohort, n=256,265 |
|---|---|---|
| Sex, men, n (%) | 401,433 (57) | 145,753 (57) |
| Age, yr | 65 (57–74) | 61 (52–70) |
| Current smoking, n (%) | 110,630 (16) | 57,702 (23) |
| Duration of diabetes mellitus, yr | 2 (0–7) | 0 (0–0) |
| Incident type 2 diabetes, n (%) | 229,635 (32) | 256,265 (100) |
| Insulin treatment, n (%) | 133,661 (19) | 25,227 (10) |
| History of cardiovascular disease, n (%) | 155,806 (22) | 43,012 (17) |
| eGFR, ml/min per 1.73 m2 | 85 (68–97) | 83 (68–96) |
| Moderate albuminuria, n (%) | 104,227 (15) | 49,536 (19) |
| Severe albuminuria, n (%) | 43,454 (6) | 5,353 (2) |
| Systolic BP, mm Hg | 138 (126–150) | 135 (124–144) |
| Body mass index, kg/m2 | 29 (26–33) | 31 (28–36) |
| HbA1c, % | 6.7 (6.2–7.6) | 6.9 (6.3–7.9) |
| Non-HDL cholesterol, mg/dl | 139 (112–170) | 127 (100–162) |
| Prescribed RASi medication, n (%) | 299,559 (42) | 38,769 (15) |
Variables are displayed as median (interquartile range) for continuous variables and count (percentage) for categorical variables. HbA1c, hemoglobin A1c; RASi, renin-angiotensin-system inhibition.
Prediction Model and Validation
Supplemental Table 4 shows the HRs and 95% confidence intervals (95% CIs) for functions A and B. The formulae for calculating survival for 3-month intervals, including coefficients and age-specific baseline hazards, are included in Supplemental Tables 5 and 6.
Predicted 10-year risks for kidney failure and all-cause mortality showed good agreement with the 10-year observed risks in the internal validation dataset (Figure 1). Internal model performance in terms of discrimination was good, reflected in c statistics of 0.89 (95% CI, 0.88 to 0.90) for kidney failure and 0.77 (95% CI, 0.77 to 0.77) for all-cause mortality.
Figure 1.
Internal validation of the prediction tool. Calibration plots for internal validation in a random sample of 25% from the Swedish National Diabetes Register (n=170,114). The calibration slope for kidney failure as the outcome was 1.02; the slope for all-cause mortality as the outcome was 1.04. (A) Kidney failure. (B) All-cause mortality. 95% CI, 95% confidence interval.
Incidence rates for kidney failure and all-cause mortality were higher in NDR than in SCI–Diabetes (Table 2). Because of the difference in event rates, the model was recalibrated according to predicted versus observed kidney failure and all-cause mortality rates. Predicted 10-year risks for kidney failure and all-cause mortality showed good agreement with the 10-year observed risks in SCI–Diabetes (Figure 2), although risk in the highest decile was overestimated. The model performed well regarding discrimination, with c statistics of 0.74 (95% CI, 0.73 to 0.76) for kidney failure and 0.77 (95% CI, 0.77 to 0.77) for all-cause mortality. A table of baseline characteristics stratified for incident versus prevalent type 2 diabetes in NDR is provided as Supplemental Table 7. A baseline table for predictors stratified according to predicted kidney failure risk is provided in Supplemental Table 8.
Table 2.
Outcomes and results for the Swedish National Diabetes Register and the Scottish Care Information–Diabetes database
| Outcomes | Swedish National Diabetes Register, n=707,077 | Scottish Care Information–Diabetes Cohort, n=256,265 |
|---|---|---|
| Median follow-up (IQR), yr | 6.8 (3.2–10.6) | 5.9 (2.6–9.6) |
| Kidney failure events, n (%) | 8004 (1.1) | 1653 (0.7) |
| All-cause mortality events, n (%) | 202,078 (29) | 45,056 (18) |
| Incidence rate, kidney failure | 1.6/1000 person-yr | 1.0/1000 person-yr |
| Incidence rate, all-cause mortality | 39.5/1000 person-yr | 28.0/1000 person-yr |
| c statistic for kidney failure | 0.89 (0.88–0.90) | 0.74 (0.73–0.76) |
| c statistic for all-cause mortality | 0.77 (0.77–0.77) | 0.77 (0.77–0.77) |
For individuals with incident type 2 diabetes in the Swedish National Diabetes Register cohort (n=229,635; 32%), incidence rates were 0.7/1000 person-years for kidney failure and 27/1000 person-years for all-cause mortality. IQR, interquartile range.
Figure 2.
External validation of the prediction tool. Calibration plots for external validation in the Scottish Care Information–Diabetes cohort (n=256,265). The calibration slope for kidney failure as the outcome was 0.73; the slope for all-cause mortality as the outcome was 0.99 after recalibration. (A) Kidney failure. (B) All-cause mortality. 95% CI, 95% confidence interval.
Individual Lifetime Estimation of Risk and Treatment Effects
An interactive user-friendly calculator is provided in Supplemental Material and will be available at https://u-prevent.com/. Individual effects from medication initiation can be modeled in terms of kidney failure–free life years gained and absolute risk reduction. Figure 3 illustrates kidney failure–free life expectancy and 10-year kidney failure risk for two individual examples with and without initiation of preventive medication. An example of a lifetable is provided in Supplemental Table 9.
Figure 3.
Example of 10-year kidney failure risk, life years free of kidney failure, and benefit from preventive treatment in two patient scenarios. The effects of initiation of renin-angiotensin-system inhibitors and sodium-glucose cotransporter-2 inhibitors on kidney failure–free lifetime expectancy are shown for two patient examples. Patient A is a 50-year-old man who is a nonsmoker with 5 years of diabetes duration, no history of cardiovascular disease, no insulin use, systolic BP of 140 mm Hg, body mass index (BMI) of 33 kg/m2, eGFR of 60 ml/min per 1.73 m2, moderate albuminuria, non-HDL cholesterol of 116 mg/dl, and hemoglobin A1c (HbA1c) of 8%. Patient B is a 65-year-old woman who is a nonsmoker with 2 years of diabetes duration, a history of cardiovascular disease, no insulin use, systolic BP of 150 mm Hg, BMI of 25 kg/m2, eGFR of 50 ml/min per 1.73 m2, severe albuminuria, non-HDL cholesterol of 155 mg/dl, and HbA1c of 8%.
Sensitivity Analyses
When incorporating the natural decline of eGFR in the predictions of kidney failure risk, model performance did not improve for 10-year predictions (Supplemental Material).
Discussion
This study describes the development and external validation of a prediction model for the estimation of 10-year and lifetime risks of kidney failure using data from almost 1 million individuals with type 2 diabetes. Furthermore, the model allows for the estimation of the individual benefit of treatment with medication most often used for kidney protection in individuals with type 2 diabetes expressed as life years gained free of kidney failure with treatment initiation. The prediction tool is available in Supplemental Material and will be provided as a calculator at https://u-prevent.com/ to allow for use in clinical practice.
Existing kidney failure prediction models developed in individuals with type 2 diabetes are on the basis of shorter prediction horizons of up to 8 years (9–11,26–28). These shorter-term predictions remain relevant for use in some patient groups (i.e., those already having advanced kidney damage) for intensifying follow-up and timing of kidney replacement therapy (29). However, for patients with lower short-term risk, including younger patients, longer-term predictions will be valuable to support decisions about preventive treatment. All models failed to adjust for competing risks. This is critical to avoid overestimating predicted kidney failure risks and treatment effects (30), especially in older individuals and individuals at low risk for kidney failure who are likely to die before developing kidney failure. Furthermore, only two previous kidney failure risk prediction models in individuals with type 2 diabetes performed external validation. Elley et al. (11) performed external validation for the 5-year risk of kidney failure in 5877 individuals with type 2 diabetes arising from the same geographic region as the derivation cohort, with a c statistic of 0.89 and reasonable calibration. Basu et al. (27) performed external validation for the 10-year risk of kidney failure in 1018 individuals with type 2 diabetes, with a c statistic for kidney failure of 0.54, and they did not perform calibration of this specific outcome. In this model, c statistics dropped from 0.89 for internal validation to 0.74 for external validation. The lower discrimination ability in the external validation is likely due to the categorical definitions of albuminuria used rather than continuous data, which may provide a better predictor, as well as the lower availability of albuminuria in the validation cohort (54% missing data). Also, diabetes duration is a relevant predictor in NDR (because this was a cohort with both prevalent and incident type 2 diabetes); however, it is not in SCI–Diabetes (because this was a cohort with incident type 2 diabetes).
In this study, the event rates for both kidney failure and all-cause mortality in individuals with type 2 diabetes were higher in Sweden compared with Scotland. The difference in kidney failure event rates is likely explained by the use of an incident cohort from SCI–Diabetes with individuals who were almost 5 years younger at cohort entry than the population of individuals with prevalent and incident diabetes identified from NDR, despite the potential for survival bias in the NDR cohort. More individuals in NDR had severe albuminuria and a history of cardiovascular disease, and the prevalence of treatment with insulin was higher. Moreover, the prevalence of renin-angiotensin-system inhibition (RASi) medication prescription was higher in NDR. This may be due to differences in antihypertensive treatment algorithms, with a more prominent role for RASi treatment in Swedish guidelines as compared with Scottish guidelines (31,32). Furthermore, because SCI–Diabetes was a cohort with incident type 2 diabetes, prescription of RASi medication is likely to have increased after diagnosis (33). Future validation and recalibration of the model will be valuable as data on individuals with type 2 diabetes with sufficient follow-up accrue to also account for differences in baseline risk due to changing patterns of medication use.
This model is intended for use in clinical practice to assess kidney failure risk in individuals with type 2 diabetes as well as likely benefits from preventive treatment. The model is underpinned by two very large and contemporary type 2 diabetes population-based cohorts with limited selection of participants. Large databases with extensive follow-up are important in order to ensure sufficient power with an adequate amount of kidney failure events because the incidence of kidney failure is relatively low as compared with cardiovascular outcomes and mortality in these populations. In external validation of this model, a slight overestimation of kidney failure risk for patients at the highest risk of kidney failure was observed, which could indicate a modest degree of overfitting in the highest-risk group. However, in clinical practice, this is unlikely to lead to erroneous decisions regarding treatment, as the true observed risk in these patients is still high and justifies intensive medical therapy. The model was developed for the entire range of eGFR. Individuals with type 2 diabetes and CKD stage 3 or 4 are likely already managed as high risk, with preventive treatment indicated. However, also in these groups, progression of kidney function decline may take several years, and the model can still act as a suitable tool to aid adherence and shared decision making in the prevention of kidney failure.
This model emphasizes lifetime benefit from treatment, which may support the initiation of preventive treatment if absolute benefit is deemed appropriate. On the contrary, the model may support not starting or postponement of preventive drug treatment if the absolute benefit is too low, and a focus on lifestyle changes might be a more appropriate initial treatment choice. In this way, lifetime risk predictions inform shared decision making while lowering the risk of side effects and polypharmacy. Furthermore, trials are often not powered to detect an effect on kidney failure risk, and albuminuria, eGFR slopes, or a combined kidney event are often used as proxies for hard kidney outcomes (34). With lifetime predictions for kidney failure, a better alternative for translating absolute kidney failure risk reduction with initiation of preventive treatment is provided.
We chose to also incorporate the effect on all-cause mortality of treatment initiation where there was substantial evidence for this because this leads to longer life expectancy and, thus, more years to develop kidney failure. However, it should be noted that kidney failure–free life years gained in individuals with a low risk of kidney failure is mostly derived from the effect on all-cause mortality. Treatment should always be considered and initiated according to current guidelines (35,36), and the kidney failure prediction tool can help support these decisions. It should further be emphasized that preventive treatment in individuals with type 2 diabetes might be initiated for other reasons than prevention or postponement of kidney failure (e.g., prevention of cardiovascular outcomes or heart failure) that were not incorporated into this model. The model therefore underestimates the total benefit of treatment. Ideally, the model should be combined with models predicting the risk of cardiovascular disease to fully capture treatment benefit (37).
The model assumes that predictors follow a natural course over time that matches the course of predictors in the derivation cohort, and model predictions are on the basis of the current predictor levels of a patient. However, follow-up in the derivation cohort was not sufficient to incorporate the natural course of predictors over the entire lifetime span, which might be particularly important for eGFR as a strong predictor for kidney failure that is known to decline with increasing age. The different methodologic approaches that we used to account for this general eGFR decline with age (e.g., incorporating standardized annual eGFR decline and modeled decline) did not improve model performance. Furthermore, the model assumes that other baseline risk factor levels follow the natural course captured in the dataset, which might not always be appropriate. However, previous studies have validated methods of estimating lifetime predictions for up to 17 years (19). Because all risk factors are subject to change after baseline and because of the general decline of eGFR with increasing age, lifetime estimations should be repeated when decisions about new treatment approaches are required.
Potential limitations of the study merit consideration. Internal and external validation was performed for 10-year risk as it is not possible to perform validation over an individual’s lifetime. Also, diabetes duration is calculated as time since diabetes diagnosis, which is unlikely to be fully accurate as some people are likely to have developed diabetes some time before a clinical diagnosis is made.
We did not have information on ethnicity, so we were not able to include this as predictor in the model. It is possible that the use of ICD-10 codes to identify outcomes may have resulted in misclassification, particularly underestimation of sustained eGFR of <15 ml/min per 1.73 m2 in the absence of long-term dialysis or transplantation as reported in a previous study (38). It is not possible to validate the ICD codes in the study populations used for this analysis or to estimate the likely effect of misclassification on the estimated discrimination and calibration of the risk models without knowing whether the degree of misclassification varies with different levels of risk factors.
We performed single imputation due to computational feasibility, which might slightly underestimate the true variability of outcome measures as opposed to multiple imputation. However, no conclusions are drawn on the basis of the significance of the model’s coefficients. Also, we chose for a split-sample approach for model development, whereas resampling methods would have been preferred. Model development was, however, still performed in >500,000 individuals with type 2 diabetes, making the power of the study more than sufficient. Another assumption made is full adherence to preventive treatment for the remaining lifetime. However, because lack of adherence is a common problem, this model might be used in aiding communication and addressing the importance of adherence to preventive treatment. Because kidney failure is a rare outcome and studies are often underpowered, treatment effects for glucose lowering and BP lowering were estimated using the best available evidence and should be interpreted with this in mind. Further research is needed to investigate to what extent the model is used in clinical practice and whether its use improves outcomes.
In conclusion, 10-year and lifetime risks of kidney failure as well as kidney failure–free life expectancy and life years free of kidney failure gained with treatment initiation can be estimated for individuals with type 2 diabetes using readily available characteristics. Assessment of individual risk and gain from treatment facilitates personalized medicine and shared decision in the management of long-term outcomes in clinical practice.
Disclosures
B. Eliasson reports consultancy agreements with Eli Lilly and Sanofi; honoraria from AstraZeneca, Eli Lilly, Novo Nordisk, and Sanofi; and speakers bureau for AstraZeneca, Novo Nordisk, and Sanofi. B. Eliasson reports personal fees from Amgen, AstraZeneca, Boehringer Ingelheim, Eli Lilly, Merck Sharp & Dohme, Mundipharma, Navamedic, Novo Nordisk, and RLS Global and grants and personal fees from Sanofi, all outside the submitted work. B. Eliasson is supported by the “Konung Gustaf V:s och Drottning Victorias Frimurarestiftelse.” S. Franzén is an employee of AstraZeneca as of October 4, 2021 and reports ownership interest in AstraZeneca. N. Halbesma is supported by British Heart Foundation Intermediate Basic Science Research Fellowship FS/16/36/32205. S. Read reports employment with, consultancy agreements with, ownership interest in, and research funding from Certara. N. Sattar reports consultancy agreements with Abbott Laboratories, Afimmune, Amgen, AstraZeneca, Boehringer Ingelheim, Eli Lilly, Hanmi Pharmaceuticals, Merck Sharp & Dohme, Novartis, Novo Nordisk, Pfizer, and Sanofi; research funding from AstraZeneca, Boehringer Ingelheim, Novartis, and Roche Diagnostics; honoraria from Abbott Laboratories, AstraZeneca, Boehringer Ingelheim, Eli Lilly, Janssen, and Novo Nordisk; serving as an associate editor for Circulation and Diabetes Care; serving as an editorial board member of Diabetes & Metabolic Syndrome: Clinical Research and Reviews; serving in an advisory or leadership role for Diabetes UK committees and the European Society of Cardiology Guideline Committee; and serving as a member of the World Heart Federation and the World Obesity Federation. All remaining authors have nothing to disclose.
Funding
None.
Supplementary Material
Acknowledgments
We thank Ann-Marie Svensson (recently deceased) for her keen interest and consistent encouragement during this work. For the Swedish NDR, we thank all of the clinicians who were involved in the care of patients with diabetes for collecting data, and we thank the staff at the Swedish NDR. We acknowledge the contributions of the people and organizations involved in providing data from, setting up, maintaining, and overseeing SCI–Diabetes, including the Scottish Diabetes Research Network supported by the National Health Service (NHS) Research Scotland, a partnership involving Scottish NHS boards and the Chief Scientist Office of the Scottish Government.
Footnotes
Published online ahead of print. Publication date available at www.cjasn.org.
Author Contributions
J.A.N. Dorresteijn, B. Eliasson, S. Franzén, N. Halbesma, H.B. Østergaard, J. van der Leeuw, F.L.J. Visseren, J. Westerink, and S.H. Wild conceptualized the study; B. Eliasson, S. Franzén, S. Read, and S.H. Wild were responsible for data curation; H.B. Østergaard and S. Read were responsible for formal analysis; J.A.N. Dorresteijn, B. Eliasson, S. Franzén, N. Halbesma, H.B. Østergaard, S. Read, N. Sattar, J. van der Leeuw, F.L.J. Visseren, J. Westerink, and S.H. Wild were responsible for methodology; H.B. Østergaard and F.L.J. Visseren were responsible for project administration; H.B. Østergaard and S. Read were responsible for validation; J. van der Leeuw provided supervision; H.B. Østergaard wrote the original draft; and J.A.N. Dorresteijn, B. Eliasson, S. Franzén, N. Halbesma, S. Read, N. Sattar, J. van der Leeuw, F.L.J. Visseren, J. Westerink, and S.H. Wild reviewed and edited the manuscript.
Data Sharing Statement
The data from the local registries are not compliant with publishing individual data in an open access institutional repository or as supporting information files with the published paper.
Supplemental Material
This article contains supplemental material online at http://cjasn.asnjournals.org/lookup/suppl/doi:10.2215/CJN.05020422/-/DCSupplemental.
Supplemental Figure 1. Smoothing and interpolation of baseline hazards.
Supplemental Figure 2. Selection of the cohort Swedish National Diabetes Register.
Supplemental Material. Predictors and missing data, statistical analyses, relative treatment effects to estimate lifelong treatment benefit in kidney failure–free life years gained, and sensitivity analyses.
Supplemental Table 1. Definition of type 2 diabetes and history of cardiovascular disease in the Swedish National Diabetes Register and the Scottish Care Information–Diabetes Database.
Supplemental Table 2. Definition of kidney failure and all-cause mortality outcomes in the Swedish National Diabetes Register and the Scottish Care Information–Diabetes Database.
Supplemental Table 3. Model assumptions.
Supplemental Table 4. Hazard ratios and 95% confidence intervals derived from multivariable Cox proportional hazard models.
Supplemental Table 5. Calculation formulas of 3-month interval survivals.
Supplemental Table 6. Age-specific baseline survival per 3-month interval.
Supplemental Table 7. Baseline characteristics of participants from the Swedish National Diabetes Register stratified according to incident or prevalent type 2 diabetes after imputation of missing data.
Supplemental Table 8. Baseline variables in the internal validation dataset (n=170,114) stratified according to predicted 10-year kidney failure risk.
Supplemental Table 9. Example of a life table.
References
- 1.Narayan KM, Boyle JP, Thompson TJ, Sorensen SW, Williamson DF: Lifetime risk for diabetes mellitus in the United States. JAMA 290: 1884–1890, 2003 [DOI] [PubMed] [Google Scholar]
- 2.Kramer A, Boenink R, Stel VS, Santiuste de Pablos C, Tomović F, Golan E, Kerschbaum J, Seyahi N, Ioanou K, Beltrán P, Zurriaga O, Magaz Á, Slon Roblero MF, Gjorgjievski N, Garneata L, Arribas F, Galvão AA, Bell S, Ots-Rosenberg M, Muñoz-Terol JM, Winzeler R, Hommel K, Åsberg A, Spustova V, Palencia García MÁ, Vazelov E, Finne P, Ten Dam MAGJ, Lopot F, Trujillo-Alemán S, Lassalle M, Kolesnyk MO, Santhakumaran S, Idrizi A, Andrusev A, Comas Farnés J, Komissarov K, Resić H, Palsson R, Kuzema V, Garcia Bazaga MA, Ziginskiene E, Stendahl M, Bonthuis M, Massy ZA, Jager KJ: The ERA-EDTA Registry Annual Report 2018: A summary. Clin Kidney J 14: 107–123, 2020 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Xia J, Wang L, Ma Z, Zhong L, Wang Y, Gao Y, He L, Su X: Cigarette smoking and chronic kidney disease in the general population: A systematic review and meta-analysis of prospective cohort studies. Nephrol Dial Transplant 32: 475–487, 2017 [DOI] [PubMed] [Google Scholar]
- 4.Zoungas S, Arima H, Gerstein HC, Holman RR, Woodward M, Reaven P, Hayward RA, Craven T, Coleman RL, Chalmers J; Collaborators on Trials of Lowering Glucose (CONTROL) group : Effects of intensive glucose control on microvascular outcomes in patients with type 2 diabetes: A meta-analysis of individual participant data from randomised controlled trials. Lancet Diabetes Endocrinol 5: 431–437, 2017 [DOI] [PubMed] [Google Scholar]
- 5.Ettehad D, Emdin CA, Kiran A, Anderson SG, Callender T, Emberson J, Chalmers J, Rodgers A, Rahimi K: Blood pressure lowering for prevention of cardiovascular disease and death: A systematic review and meta-analysis. Lancet 387: 957–967, 2016 [DOI] [PubMed] [Google Scholar]
- 6.Catalá-López F, Macías Saint-Gerons D, González-Bermejo D, Rosano GM, Davis BR, Ridao M, Zaragoza A, Montero-Corominas D, Tobías A, de la Fuente-Honrubia C, Tabarés-Seisdedos R, Hutton B: Cardiovascular and renal outcomes of renin-angiotensin system blockade in adult patients with diabetes mellitus: A systematic review with network meta-analyses. PLoS Med 13: e1001971, 2016 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Palmer SC, Tendal B, Mustafa RA, Vandvik PO, Li S, Hao Q, Tunnicliffe D, Ruospo M, Natale P, Saglimbene V, Nicolucci A, Johnson DW, Tonelli M, Rossi MC, Badve SV, Cho Y, Nadeau-Fredette AC, Burke M, Faruque LI, Lloyd A, Ahmad N, Liu Y, Tiv S, Millard T, Gagliardi L, Kolanu N, Barmanray RD, McMorrow R, Raygoza Cortez AK, White H, Chen X, Zhou X, Liu J, Rodríguez AF, González-Colmenero AD, Wang Y, Li L, Sutanto S, Solis RC, Díaz González-Colmenero F, Rodriguez-Gutierrez R, Walsh M, Guyatt G, Strippoli GFM: Sodium-glucose cotransporter protein-2 (SGLT-2) inhibitors and glucagon-like peptide-1 (GLP-1) receptor agonists for type 2 diabetes: Systematic review and network meta-analysis of randomised controlled trials. BMJ 372: m4573, 2021 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Slieker RC, van der Heijden AAWA, Siddiqui MK, Langendoen-Gort M, Nijpels G, Herings R, Feenstra TL, Moons KGM, Bell S, Elders PJ, M’t Hart L, Beulens JWJ: Performance of prediction models for nephropathy in people with type 2 diabetes: Systematic review and external validation study. BMJ 374: n2134, 2021 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Wan EYF, Fong DYT, Fung CSC, Yu EYT, Chin WY, Chan AKC, Lam CLK: Prediction of new onset of end stage renal disease in Chinese patients with type 2 diabetes mellitus - A population-based retrospective cohort study. BMC Nephrol 18: 257, 2017 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Lin CC, Li CI, Liu CS, Lin WY, Lin CH, Yang SY, Li TC: Development and validation of a risk prediction model for end-stage renal disease in patients with type 2 diabetes. Sci Rep 7: 10177, 2017 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Elley CR, Robinson T, Moyes SA, Kenealy T, Collins J, Robinson E, Orr-Walker B, Drury PL: Derivation and validation of a renal risk score for people with type 2 diabetes. Diabetes Care 36: 3113–3120, 2013 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Matsushita K, van der Velde M, Astor BC, Woodward M, Levey AS, de Jong PE, Coresh J, Gansevoort RT; Chronic Kidney Disease Prognosis Consortium : Association of estimated glomerular filtration rate and albuminuria with all-cause and cardiovascular mortality in general population cohorts: A collaborative meta-analysis. Lancet 375: 2073–2081, 2010 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Gudbjörnsdottir S, Cederholm J, Nilsson PM, Eliasson B; Steering Committee of the Swedish National Diabetes Register : The National Diabetes Register in Sweden: An implementation of the St. Vincent Declaration for Quality Improvement in Diabetes Care. Diabetes Care 26: 1270–1276, 2003 [DOI] [PubMed] [Google Scholar]
- 14.Cunningham S, McAlpine R, Leese G, Brennan G, Sullivan F, Connacher A, Waller A, Boyle DI, Greene S, Wilson E, Emslie-Smith A, Morris AD: Using web technology to support population-based diabetes care. J Diabetes Sci Technol 5: 523–534, 2011 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Levey AS, Eckardt KU, Tsukamoto Y, Levin A, Coresh J, Rossert J, De Zeeuw D, Hostetter TH, Lameire N, Eknoyan G: Definition and classification of chronic kidney disease: A position statement from Kidney Disease: Improving Global Outcomes (KDIGO). Kidney Int 67: 2089–2100, 2005 [DOI] [PubMed] [Google Scholar]
- 16.Steyerberg EW: Clinical Prediction Models: A Practical Approach to Development, Validation and Updating, New York, Springer, 2009 [Google Scholar]
- 17.Levey AS, Stevens LA, Schmid CH, Zhang YL, Castro AF 3rd, Feldman HI, Kusek JW, Eggers P, Van Lente F, Greene T, Coresh J; CKD-EPI (Chronic Kidney Disease Epidemiology Collaboration) : A new equation to estimate glomerular filtration rate. Ann Intern Med 150: 604–612, 2009 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Arsenault BJ, Rana JS, Stroes ES, Després JP, Shah PK, Kastelein JJ, Wareham NJ, Boekholdt SM, Khaw KT: Beyond low-density lipoprotein cholesterol: Respective contributions of non-high-density lipoprotein cholesterol levels, triglycerides, and the total cholesterol/high-density lipoprotein cholesterol ratio to coronary heart disease risk in apparently healthy men and women. J Am Coll Cardiol 55: 35–41, 2009 [DOI] [PubMed] [Google Scholar]
- 19.Dorresteijn JA, Kaasenbrood L, Cook NR, van Kruijsdijk RC, van der Graaf Y, Visseren FL, Ridker PM: How to translate clinical trial results into gain in healthy life expectancy for individual patients. BMJ 352: i1548, 2016 [DOI] [PubMed] [Google Scholar]
- 20.Harrell FE Jr, Califf RM, Pryor DB, Lee KL, Rosati RA: Evaluating the yield of medical tests. JAMA 247: 2543–2546, 1982 [PubMed] [Google Scholar]
- 21.Wolff RF, Moons KGM, Riley RD, Whiting PF, Westwood M, Collins GS, Reitsma JB, Kleijnen J, Mallett S; PROBAST Group : PROBAST: A tool to assess the risk of bias and applicability of prediction model studies. Ann Intern Med 170: 51–58, 2019 [DOI] [PubMed] [Google Scholar]
- 22.Collins GS, Reitsma JB, Altman DG, Moons KG: Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): The TRIPOD statement. BMJ 350: g7594, 2015 [DOI] [PubMed] [Google Scholar]
- 23.Gellert C, Schöttker B, Brenner H: Smoking and all-cause mortality in older people: Systematic review and meta-analysis. Arch Intern Med 172: 837–844, 2012 [DOI] [PubMed] [Google Scholar]
- 24.McGuire DK, Shih WJ, Cosentino F, Charbonnel B, Cherney DZI, Dagogo-Jack S, Pratley R, Greenberg M, Wang S, Huyck S, Gantz I, Terra SG, Masiukiewicz U, Cannon CP: Association of SGLT2 inhibitors with cardiovascular and kidney outcomes in patients with type 2 diabetes: A meta-analysis. JAMA Cardiol 6: 148–158, 2021 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Sattar N, Lee MMY, Kristensen SL, Branch KRH, Del Prato S, Khurmi NS, Lam CSP, Lopes RD, McMurray JJV, Pratley RE, Rosenstock J, Gerstein HC: Cardiovascular, mortality, and kidney outcomes with GLP-1 receptor agonists in patients with type 2 diabetes: A systematic review and meta-analysis of randomised trials. Lancet Diabetes Endocrinol 9: 653–662, 2021 [DOI] [PubMed] [Google Scholar]
- 26.Wysham CH, Gauthier-Loiselle M, Bailey RA, Manceur AM, Lefebvre P, Greenberg M, Duh MS, Young JB: Development of risk models for major adverse chronic renal outcomes among patients with type 2 diabetes mellitus using insurance claims: A retrospective observational study. Curr Med Res Opin 36: 219–227, 2020 [DOI] [PubMed] [Google Scholar]
- 27.Basu S, Sussman JB, Berkowitz SA, Hayward RA, Yudkin JS: Development and validation of Risk Equations for Complications Of type 2 Diabetes (RECODe) using individual participant data from randomised trials. Lancet Diabetes Endocrinol 5: 788–798, 2017 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Hayes AJ, Leal J, Gray AM, Holman RR, Clarke PM: UKPDS outcomes model 2: A new version of a model to simulate lifetime health outcomes of patients with type 2 diabetes mellitus using data from the 30 year United Kingdom Prospective Diabetes Study: UKPDS 82. Diabetologia 56: 1925–1933, 2013 [DOI] [PubMed] [Google Scholar]
- 29.Eckardt KU, Bansal N, Coresh J, Evans M, Grams ME, Herzog CA, James MT, Heerspink HJL, Pollock CA, Stevens PE, Tamura MK, Tonelli MA, Wheeler DC, Winkelmayer WC, Cheung M, Hemmelgarn BR; Conference Participants : Improving the prognosis of patients with severely decreased glomerular filtration rate (CKD G4+): Conclusions from a Kidney Disease: Improving Global Outcomes (KDIGO) Controversies Conference. Kidney Int 93: 1281–1292, 2018 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Wolbers M, Koller MT, Witteman JC, Steyerberg EW: Prognostic models with competing risks: Methods and application to coronary risk prediction. Epidemiology 20: 555–561, 2009 [DOI] [PubMed] [Google Scholar]
- 31.Landstinget Sörmland Lakemedelskommitten : Behandlingsriktlinjer hypertoni, 2016. Available at: https://samverkan.regionsormland.se/siteassets/halsa-och-vard/lakemedelskommitten/behandlingsriktlinjer/behandlingsriktllinjer_hypertoni_2016.pdf. Accessed February 17, 2022
- 32.Health Improvement Scotland : Risk estimation and the prevention of cardiovascular disease, 2017. Available at: https://www.sign.ac.uk/our-guidelines/risk-estimation-and-the-prevention-of-cardiovascular-disease/. Accessed February 17, 2022
- 33.Wright AK, Suarez-Ortegon MF, Read SH, Kontopantelis E, Buchan I, Emsley R, Sattar N, Ashcroft DM, Wild SH, Rutter MK: Risk factor control and cardiovascular event risk in people with type 2 diabetes in primary and secondary prevention settings. Circulation 142: 1925–1936, 2020 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Greene T, Ying J, Vonesh EF, Tighiouart H, Levey AS, Coresh J, Herrick JS, Imai E, Jafar TH, Maes BD, Perrone RD, Del Vecchio L, Wetzels JFM, Heerspink HJL, Inker LA: Performance of GFR slope as a surrogate end point for kidney disease progression in clinical trials: A statistical simulation. J Am Soc Nephrol 30: 1756–1769, 2019 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Visseren FLJ, Mach F, Smulders YM, Carballo D, Koskinas KC, Bäck M, Benetos A, Biffi A, Boavida JM, Capodanno D, Cosyns B, Crawford C, Davos CH, Desormais I, Di Angelantonio E, Franco OH, Halvorsen S, Hobbs FDR, Hollander M, Jankowska EA, Michal M, Sacco S, Sattar N, Tokgozoglu L, Tonstad S, Tsioufis KP, van Dis I, van Gelder IC, Wanner C, Williams B; ESC National Cardiac Societies; ESC Scientific Document Group : 2021 ESC guidelines on cardiovascular disease prevention in clinical practice. Eur Heart J 42: 3227–3337, 2021 [DOI] [PubMed] [Google Scholar]
- 36.Kidney Disease: Improving Global Outcomes (KDIGO) Diabetes Work Group : KDIGO 2020 clinical practice guideline for diabetes management in chronic kidney disease. Available at: https://kdigo.org/wp-content/uploads/2020/10/KDIGO-2020-Diabetes-in-CKD-GL.pdf. Accessed November 23, 2021 [Google Scholar]
- 37.Berkelmans GFN, Gudbjörnsdottir S, Visseren FLJ, Wild SH, Franzen S, Chalmers J, Davis BR, Poulter NR, Spijkerman AM, Woodward M, Pressel SL, Gupta AK, van der Schouw YT, Svensson AM, van der Graaf Y, Read SH, Eliasson B, Dorresteijn JAN: Prediction of individual life-years gained without cardiovascular events from lipid, blood pressure, glucose, and aspirin treatment based on data of more than 500 000 patients with type 2 diabetes mellitus. Eur Heart J 40: 2899–2906, 2019 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Norton JM, Grunwald L, Banaag A, Olsen C, Narva AS, Marks E, Koehlmoos TP: CKD prevalence in the Military Health System: Coded versus uncoded CKD. Kidney Med 3: 586–595.e1, 2021 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
