Prognostic Value of Claims Data for Chronic Kidney Disease-Associated Health Risks Authors: Sebastian Schwab, Volquart Stoy, Olga Dortmann, Janika Raudzus, Thomas Mayrhofer, Sibylle von Vietinghoff Published: 2026 ABSTRACT AbstractBackgroundChronic kidney disease (CKD) associates with kidney failure and overall health risks. We here investigated the prognostic value of claims data in the absence of biochemical data.MethodsA total of 29,144 persons insured at AOK Rhineland-Hamburg with a diagnosis of CKDIII or CKDIV in 2016 were included in analyses of renal replacement therapy (rRT), death, hospital admissions and diagnoses, and prescription of CKD-related therapies during the ensuing 6-year period (2017–2022) using corresponding billing codes.ResultsIn 22,228 individuals diagnosed with CKDIII and 6,916 with CKDIV during the base year, rRT was prescribed in 3.8% and 14.3% during the observation period; 7.0% and 13.3% were admitted to the hospital for acute kidney injury (AKI). A combination of claims data significantly contributed to the prognosis of rRT (AUC 0.87) and AKI (AUC 0.68) and overall outcomes, death (AUC 0.97), hospital admission for heart failure (AUC 0.76), and intensive care unit (ICU) therapy (AUC 0.67). All AUCs significantly improved after inclusion of CKD stage into the model. Outpatient prescriptions of renin-angiotensin system and sodium-glucose cotransporter 2 (SGLT2) inhibitors are associated with significantly better renal and cardiovascular outcomes. Prescription of CKD-related pharmacotherapy for calcium metabolism (AUC 0.74), hyperkalemia/hyperphosphatemia (AUC 0.83), metabolic acidosis (AUC 0.81), erythropoiesis-stimulating agents (AUC 0.65), and hyperuricemia therapy (AUC 0.72) was prognosticated by claims data with significant contribution of CKD stage.Summary and ConclusionClaims data provide prognostic information for renal and overall outcomes in CKDIII and CKDIV patients even in the absence of laboratory measurements. Inclusion of additional values should be evaluated in patients with and without CKD for individualized risk calibration. Introduction Chronic kidney disease (CKD) is common and conveys major risks for renal adverse events, overall survival and comorbidities, namely, in the cardiovascular system [1–3]. Laboratory measurements, particularly estimated glomerular filtration rate (eGFR) and urinary protein loss, are central predictors of CKD progression and kidney failure risk [2]. Individualized risk calibration models for kidney functional deterioration, most prominently renal replacement therapy (rRT), were developed and tested in various populations [4–6]. Beyond eGFR, urinary protein excretion is a major contributor to disease progression in individuals with advanced CKD [7], and diabetes mellitus II [8, 9] and also in respect to GFR slope [10, 11]. A combination of the eight variables age, sex, GFR, urine albumin, and calcium, phosphate, bicarbonate, and albumin developed in a Canadian cohort performed best in a wide range of settings [5]. This was recently verified [12]. However, laboratory measurements are frequently incomplete, limiting practical applicability of these models [12]. Overall health burden and costs rise with advancing CKD [2, 13]. A 2014 estimate even projected more than 10% of health care spending in Germany to individuals with CKD billing codes [14]. Multimorbidity, which also increases CKD progression risk, further contributes [15, 16]. Male sex, increasing age, and socioeconomic disadvantage are established risk factors associated with CKD progression and adverse outcomes [17–20]. On the other hand, the actual requirement for rRT within 5 years was lower for older than younger individuals stratified by CKD diagnosis using German claims data [21], possibly reflecting competing mortality risks and differences in treatment decisions in elderly populations. In addition, socioeconomic status has repeatedly been associated with CKD progression and adverse outcomes, including in recent UK population-based analyses [22]. Inclusion of such widely available data into risk scores is warranted. Cardiovascular outcomes strongly and independently associate with decreasing kidney function [1, 2]. This also applies to older adults with moderate risk of kidney disease progression and is in part independent of proteinuria [23, 24]. Efforts are ongoing to routinely include kidney functions, such as GFR and albuminuria, into cardiovascular risk prediction tools [25, 26]. On the other hand, German claims data suggest low prescription rates of indicated therapies and high mortality in CKD patients after myocardial infarction [27]. Individualizing risk assessment without the need to obtain additional data could improve access to therapies. CKD is associated with multiple metabolic and endocrine complications requiring specific pharmacologic management, including disturbances in calcium metabolism, electrolyte balance, acid-base homeostasis, anemia, and hyperuricemia [2]. All significantly add to the health and financial burden of CKD [2, 28]. Improved risk stratification approaches may facilitate earlier identification of patients at increased risk for CKD-associated complications and support timely therapeutic intervention. We here evaluated the prognostic value of claims data in patients with CKDIII and CKDIV diagnoses for renal and overall outcomes and the need for CKD-related therapies during a 6-year period. Methods Data Source Claims data of statutory health insurance AOK Rhineland-Hamburg from 2016 to 2022 were analyzed. Available claims-based variables included demographic characteristics, ICD-10 coded diagnoses, outpatient and inpatient procedures, prescription claims, healthcare utilization variables, specialist consultations, laboratory billing codes, and participation in structured disease management programs. Inclusion and Classification of Patients The year 2016 served as the base year for cohort disease severity and comorbidity definition. Prespecified follow-up events (renal replacement therapy or drug treatment for CKD-associated metabolic and endocrine complications) occurring in 2016 led to exclusion in the respective regression model. Events and therapeutic and surveillance measures were recorded during the observation period (2017–2022). Target variables and comorbidities were recorded using coded World Health Organization (WHO) ICD-10 diagnoses [29]. Outpatient diagnoses were considered present if they were coded in at least 2 quarters or supplemented by specific medication. All inpatient diagnoses were considered. Medication was recorded using the Anatomical Therapeutic Chemical Classification System (ATC codes) [30]. Medication supply was operationalized using medication possession rate (MPR). Continuous supply was defined if medication was available on ≥80% of the days of the individual observation period [31]. Drug categories included diuretics (C03), angiotensin-converting enzyme (ACE) inhibitors and angiotensin II receptor blockers (AT II receptor blockers) (C09, renin-angiotensin system [RAS] blockers), and SGLT2 inhibitors (A10BK, A10BD15, A10BD19, A10BD24). EBM billing codes [32] for outpatient medical services in the statutory health insurance system in Germany assign a specific number for medical contact or procedure. OPS billing codes were employed for inpatients. Full ATC, OPS, and EBM codes are shown in online supplementary Table 1 (for all online suppl. material, see https://doi.org/10.1159/000553419). Contact with a nephrologist was assumed if it occurred at least yearly. Contacts with cardiology, rheumatology, or urology specialists and participation in health “checkup” including physician consultation, a physical examination, and basic urine and blood laboratory tests including serum creatinine and urine dipstick analysis [33] were operationalized as dichotomous variables. The Disease Management Program (DMP) for diabetes mellitus type 2 [34] consists of structured documentation of clinical parameters including HbA1c, blood pressure, body mass index, comorbidities as well as treatment, counseling, and training measures. At least one medical DMP entry in the base year or observation period was operationalized as participation. The target variables are analyzed depending on the event from prior knowledge of pathophysiology and statistical associations. The start of renal replacement therapy was recorded using ICD-10 coding. Endpoints were identified for inpatient treatments based on hospital cases with the respective main discharge diagnoses and for drug prescriptions with the relevant ATC codes as detailed in online supplement Table 1. Statistical Analysis Continuous variables are expressed as mean ± standard deviation (SD), while categorical variables are shown as absolute and relative frequencies. Comparisons utilized independent sample t tests with Welch’s correction where appropriate for continuous variables and Fisher’s exact or chi-square tests for categorical variables. Multivariable logistic regression analyses adjusting for known risk factors were performed to investigate the prognostic value of CKD on the occurrence of AKI, rRT, death, admission for heart failure, and ICU admission. Candidate variables included demographic characteristics, comorbidities, healthcare utilization measures, surveillance variables, pharmacologic therapies, and CKD stage. Full models were reduced using a backward stepwise selection approach based on a significance level of 0.05. The area under the receiver operating characteristic (ROC) curve (AUC) was used to evaluate the incremental value of CKD stage beyond other claims-based variables. Model comparisons were performed using likelihood-ratio tests. Given the exploratory nature of the study, statistical inference was guided by a two-sided 5% false positive error rate without adjustment for multiple comparisons. All statistical analyses were conducted using SAS version 9.5. Results Study Population and CKD Cohort AOK Rhineland-Hamburg health insurance covered 2.9 million individuals in 2016 (shown in Fig. 1). A total of 95,455 received at least one inpatient or at least 2 outpatient CKD diagnoses (N18) during this year. This reflects an overall prevalence of 1.4% CKDIII and 0.4% CKDIV according to these billing codes. Patients receiving renal replacement therapy by hemodialysis or peritoneal dialysis, kidney transplant recipients, and patients with incomplete diagnoses during the base year and patients without CKD diagnosis (N18) during the observation period were excluded. A total of 22,228 and 6,916 persons were specified by N18.3 or N18.4 codes as having CKDIII or CKDIV, respectively, and further analyzed. These were on average 76 ± 11 years of age, 50% female in the CKDIII, and 77 ± 11 years old, 57% female in the CKDIV group (shown in Table 1), i.e., CKDIV patients were slightly, but significantly, older (<0.0001) and more often female. Common co-diagnosed conditions were hypertension (appr. 95% in both groups) and diabetes (57% in both groups). Only a minority were diagnosed as having proteinuria (4.2% vs. 8.1%) or with specific kidney diseases such as polycystic kidney disease (4.9% vs. 6.3%) or vasculitides (0.9% vs. 1.5%). These diagnoses were more commonly associated with a CKDIV than CKDIII code. Flow diagram of patient inclusion. The insured population was screened for CKD codes. After exclusion of patients on rRT, kidney graft recipients during the base year, 22,228 persons with N18.3 and 6,916 with N18.4 code were included into the analysis. Table 1.Characteristics of total cohort and CKDIII and CKDIV patients​Total cohortCKDIII (n = 22,228)CKDIV (n = 6,916) p value (CKDIII vs. IV)Basic characteristics n (%) Age, years, (mean ± SD)76.2±11.175.9±10.977.1±11.3<0.0001*** Gender (female)51.6 (15,067)50.1 (11,137)56.8 (3,930)<0.0001***Socioeconomic status (%)​ Working4.2 (1,236)4.4 (982)3.6 (254)<0.0001*** Retired88.9 (25,931)88.8 (19,748)89.4 (6,183)0.2 Social security2.3 (681)2.3 (530)2.1 (151)0.36Renal diagnoses (%) Proteinuria5.1 (1,495)4.1 (932)8.1 (563)<0.0001*** Glomerulonephritis/vasculitis1.0 (310)0.9 (209)1.4 (101)0.0003*** Cystic kidney diseases5.2 (1,533)4.9 (1,095)6.3 (438)<0.0001***Cardiovascular profile (%) Hypertension95.2 (27,752)94.8 (21,074)96.5 (6,678)<0.0001*** Diabetes mellitus57.3 (16,718)57.2 (12,729)57.6 (3,989)0.55 Coronary artery disease51.4 (14,997)50.7 (11,269)53.9 (3,728)<0.0001*** Heart failure42.7 (12,448)40.6 (9.044)49.2 (3,404)<0.0001*** Atrial fibrillation32.9 (9,598)31,7 (7,057)36.7 (2,541)<0.0001*** Stroke17.8 (5,200)17.4 (3,883)19.0 (1,317)0.003** Peripheral artery disease31.4 (9,175)30.9 (6,868)33.3 (2,307)0.0001*** Smoking10.0 (2,932)10,1 (2,260)9.7 (672)0.28Other frequent comorbidities (%) Malignoma5.7 (1,686)5.4 (1,221)6.7 (465)0.74 COPD22.8 (6,666)22.4 (4,978)24.4 (1,688)0.0005*** Dementia15.0 (4,380)14.5 (3,241)16.5 (1,139)0.0002***Monitoring and medication (%) Albuminuria measurement5.7 (1,667)5.5 (1,238)6.3 (439)<0.0001*** Creatinine measurement60.5 (17,640)59.8 (13,300)62.7 (4,340)0.06 “Checkup” examination57.2 (16,674)59.6 (13,247)49.5 (3,427)<0.0001*** DMP diabetes33.6 (9,795)34.7 (7,721)30.0 (2,074)<0.0001*** Nephrologist consultation15.3 (4,473)10.8 (2,412)29.8 (2,061)<0.0001*** Chemotherapy10.5 (3,068)10.5 (2,348)10.4 (720)0.0002*** RAS blockade49.5 (14,438)50.7 (11,270)45.8 (3,168)<0.0001*** Diuretic therapy57.6 (16.795)54.1 (12.037)68.8 (4.758)<0.0001*** SGLT2 inhibition0.6 (199)0.8 (185)0.2 (14)<0.0001***Fisher’s exact test for numbers and T test with Welch’s correction for age. COPD: Chronic obstructive pulmonary disease DMP, Disease management program. Corresponding codes are listed in online supplement Table 1. Characteristics of total cohort and CKDIII and CKDIV patients Fisher’s exact test for numbers and T test with Welch’s correction for age. COPD: Chronic obstructive pulmonary disease DMP, Disease management program. Corresponding codes are listed in online supplement Table 1. Atherosclerotic and cardiovascular conditions during the base year also were more common in the CKDIV subpopulation, including heart failure, atrial fibrillation, peripheral arterial disease, and stroke (shown in Table 1). This also applies to chronic obstructive pulmonary disease, malignancy, and dementia. Overall, these findings indicate higher age and comorbidity burden in CKDIV compared with CKDIII-diagnosed individuals. Higher Adverse Event Rates and Prescriptions for CKD-Related Conditions in Individuals with CKDIV than CKDIII Diagnosis The cohorts were analyzed for overall and kidney-specific outcomes during the ensuing 6 years (2017–2022). During this observation period, events that necessarily elicit billing of services were considered. Observation time was 1,511 days for patients diagnosed with CKDIII and 1,301 days in patients with CKDIV (shown in Table 2). 52.3% and 66.4% of patients died, respectively. Table 2.Frequency of endpoints and prescription of supportive therapies​Total cohortCKDIIICKDIV p value (CKDIII vs. IV)Observation time (days)1,461±7881,511±7841,301±801<0.0001***Events rRT6.3% (1,838/29,013)3.8% (854/22,176)14.3% (984/6,837)<0.0001*** Death55.7% (16,255/29,144)52.4% (11,661/22,228)66.4% (4,594/6,916)<0.0001***Hospital admission Any89.24% (26,010/29,144)88.46% (19,662/22,228)91.79% (6,348/6,916)<0.0001*** AKI8.5% (2,501/29,144)7.0% (1,577/22,228)13.3% (924/6,916)<0.0001*** Heart failure25.3% (7,392/29,144)23.8% (5,298/22,228)30.2% (2,094/6,916)<0.0001*** Myocardial infarction6.4% (1,883/29,144)6.3% (1,406/22,228)6.8% (477/6,916)0.11 Stroke5.8% (1,701/29,144)5.8% (1,309/22,228)5.6% (392/6,916)0.53 ICU admission9.9% (2,901/29,144)9.3% (2,074/22,228)11.9% (827/6,916)<0.0001***Supportive therapies Calcium metabolism26.1% (6,070/23,216)23.9% (4,461/18,596)34.8% (1,609/4,620)<0.0001*** Hyperkalemia and Hyperphosphatemia6.4% (1,850/28,784)4.3% (960/22,078)13.2% (890/6,706)<0.0001*** Metabolic acidosis8.2% (2,308/27,811)6.4% (1,409/21,756)14.8% (899/6,055)<0.0001*** Erythropoiesis stimulating agents8.4% (2,421/28,565)5.7% (1,263/22,044)17.7% (1,158/6,521)<0.0001*** Blood transfusion25.0%23.3% (5,182/22,228)30.6% (2,120/6,916)<0.0001***7,302/29,144 Hyperuricemia19.9% (3,704/18,590)19.3% (2,882/14,884)22.1% (822/3,706)0.002** p values of Fisher’s exact tests are shown. Corresponding codes are listed in online supplement Table 1. Frequency of endpoints and prescription of supportive therapies p values of Fisher’s exact tests are shown. Corresponding codes are listed in online supplement Table 1. Regarding kidney outcomes, renal replacement therapy (rRT) was claimed for 3.8% of the CKDIII and 14.3% of the CKDIV population (shown in Table 2). Hospital admissions were billed for 88.4% of CKDIII and 91.7% of the CKDIV individuals. Admission for acute kidney injury (AKI) occurred in 7.0% and 13.3%, respectively. Hospital admissions for heart failure, but not for myocardial infarction or stroke, and ICU admissions were also recorded more frequently for individuals with CKDIII and CKDIV (shown in Table 2). All investigated supportive pharmacologic therapies were significantly more frequently prescribed to CKDIV than CKDIII patients (shown in Table 2). Prescriptions for calcium metabolic disorders were 31% more frequent; the relative increases for hyperkalemia and hyperphosphatemia therapies were 67%, for metabolic acidosis 57%, for erythropoiesis-stimulating agent 67%, and for hyperuricemia 13% for CKDIV compared to CKDIII individuals. Prognostic Role of Clinical and Demographic Parameters Reflected in Claims Data for Renal and Overall Outcomes We next tested the discrimination of major adverse events by parameters recorded during the base year, with and without consideration of CKD stage. Logistic regression analysis yielded significant prognostic values for major kidney events, namely, the need for rRT (AUC 0.87) and hospital admission for AKI (AUC 0.68) (shown in Fig. 2a and Table 3). The risks of rRT and admission for AKI were higher in men than women but mildly decreased with age. Retirement and the need for social security support increased both. CKD stage IV increased the risk of rRT 3.01-fold and AKI 1.65-fold compared to stage III (shown in Table 3). This is reflected by a significant increase of the AUC for both (shown in Fig. 2a). Among common comorbidities, hypertension, and diabetes increased both risks as did treatment of malignant neoplasms (Table 3). Sensitivity and specificity of outcome assessment using claims data with and without inclusion of CKD stage. a–c Main kidney outcomes of renal replacement therapy (rRT) and acute kidney injury (AKI) (a), death (b), and hospital admission for heart failure and ICU admission (c) with and without inclusion of the CKD stage into the model (full models in Tables 2 and 3, p < 0.001 for comparison of AUC with and without consideration of CKD stage). Table 3.Logistic regression analysis for subsequent acute kidney injury (AKI) and renal replacement therapy (rRT)Renal replacement therapy (rRT)Acute kidney injury (AKI)ORCI-95 p valueORCI-95 p valueObservation time1.001.00–1.00<0.001Gender FemaleRef.Ref. Male1.681.49–1.89<0.0011.231.13–1.35<0.001Age0.950.95–0.96<0.0010.990.98–0.99<0.001Socioeconomic status WorkingRef.Ref. Retired (<800€)1.351.06–1.720.0162.121.58–2.85<0.001 Retired (800–1,599€)1.210.95–1.550.1282.221.65–2.99<0.001 Retired (>1,600€)1.220.92–1.620.1742.411.76–3.30<0.001 Self-employment0.710.24–2.100.5370.490.07–3.630.484 Social security1.571.17–2.120.0032.011.39–2.91<0.001 Others1.160.85–1.570.3491.731.22–2.450.002Care level 0Ref.Ref. 10.670.40–1.110.1170.960.65–1.420.833 20.430.25–0.730.0020.840.56–1.250.389 30.260.11–0.620.0020.500.31–0.830.007Nursing care type 0 (none)Ref.Ref. 1 (nursing home)0.720.39–1.340.3021.080.71–1.640.715 2 (at home)1.250.76–2.050.3861.340.91–1.970.138Kidney conditionsCKD Stage IIIRef.<0.001Ref. IV3.012.69–3.381.651.51–1.81<0.001 Cystic kidney diseases1.331.08–1.640.007 Proteinuria1.351.14–1.60<0.001Cardiovascular profile Hypertension1.421.06–1.910.020 Diabetes1.831.59–2.10<0.0011.411.27–1.57<0.001 Heart failure1.221.11–1.34<0.001 Atrial fib1.241.12–1.38<0.001 Peripheral artery disease1.141.02–1.290.028 Smoking1.431.21–1.69<0.001Other frequent comorbidities Malignoma1.651.42–1.93<0.0011.381.22–1.56<0.001 COPD1.261.12–1.42<0.001 Polyarthropathy0.740.56–0.990.041 Dementia0.780.62–1.000.045Obesity grade NoneRef. BMI 30–341.130.98–1.300.097 BMI 35–391.010.86–1.190.873 BMI >401.311.12–1.540.0008 Not specified1.100.97–1.250.1427Medical attention and surveillance measures Albuminuria measurement (yearly)0.680.54–0.84<0.0010.690.57–0.84<0.001 Nephrologist contact (yearly)1.541.38–1.72<0.001 “Checkup” examination0.720.64–0.81<0.0010.860.79–0.94<0.001 DMP diabetes0.710.61–0.81<0.0010.730.65–0.81<0.001 Medical specialist contact0.830.75–0.92<0.001Pharmacologic therapy Diuretic therapy1.581.42–1.75<0.0012.061.86–2.28<0.001 RAS blockade0.890.81–0.980.020 SGLT2 inhibition0.360.16–0.830.016OR, odds ratio; CI-95, 95% confidence interval; Ref., reference group; Atrial fib., atrial fibrillation; COPD, chronic obstructive pulmonary disease. Only factors with significant contribution are shown. Logistic regression analysis for subsequent acute kidney injury (AKI) and renal replacement therapy (rRT) OR, odds ratio; CI-95, 95% confidence interval; Ref., reference group; Atrial fib., atrial fibrillation; COPD, chronic obstructive pulmonary disease. Only factors with significant contribution are shown. Regarding death, expectedly age, male gender, and lower socioeconomic status increased the risk together with cardiovascular, malignant, and other comorbidities (shown in Table 4). Also here, CKD stage significantly added to the risk (1.37-fold) with a significant increase in the AUC (0.97, shown in Fig. 2b). A more advanced CKD stage increased the risk of hospital admission for heart failure 1.12-fold, while persons with CKDIV were admitted to the ICU significantly less frequently (0.83-fold) (shown in Table 4). Both events are significantly associated with a combination of claims data (AUC 0.76, AUC 0.67, shown in Fig. 2c). Table 4.Logistic regression analysis for subsequent death, heart failure and admission to the intensive care unit (ICU)​DeathHeart failureICUORCI-95 p valueORCI-95 p valueORCI-95 p valueObservation time0.990.99–0.99<0.001​​​1.001.00–1.00<0.001Gender FemaleRef.​​​​​Ref.​​ Male1.291.16–1.42<0.001​​​1.201.10–1.31<0.001Age1.061.06–1.07<0.0011.011.01–1.02<0.0010.970.97–0.98<0.001Socioeconomic status WorkingRef.​​Ref.​​Ref.​​ Retired (<800€)3.512.21–5.55<0.0011.841.44–2.37<0.0012.461.88–3.22<0.001 Retired (800–1,599€)3.522.22–5.59<0.0011.891.47–2.43<0.0012.371.81–3.11<0.001 Retired (>1,600€)3.832.39–6.15<0.0011.821.41–2.36<0.0012.451.84–3.27<0.001 Self-employment3.130.71–13.850.1340.350.05–2.620.305<0.01<0.01 – >999.90.928 Social security4.262.45–7.40<0.0011.871.35–2.58<0.0011.421.00–2.030.050 Others1.370.81–2.310.2351.681.27–2.23<0.0012.031.49–2.78<0.001Care level 0Ref.​​Ref.​​Ref.​​ 11.431.28–1.61<0.0011.000.74–1.360.9960.810.56–1.160.246 21.951.63–2.34<0.0010.770.57–1.060.1060.770.53–1.120.171 33.972.42–6.52<0.0010.400.28–0.58<0.0010.470.30–0.760.002Nursing care type 0 (none)​​​Ref.​​Ref.​​ 1 (nursing home)​​​0.650.47–0.900.0100.920.62–1.380.697 2 (at home)​​​1.040.77–1.410.8141.491.04–2.130.029Kidney conditionsCKD stage IIIRef.​​Ref.​​Ref.​​ IV1.371.22–1.53<0.0011.121.05–1.190.0010.830.76–0.91<0.001Cardiovascular profile Hypertension1.311.01–1.690.044​​​​​​ Diabetes1.371.24–1.51<0.0011.341.24–1.44<0.0011.481.34–1.64<0.001 Heart failure1.121.01–1.250.0281.711.61–1.83<0.001​​​ CAD​​​1.361.28–1.45<0.0011.111.02–1.210.021 Atrial fib1.241.12–1.38<0.0011.791.68–1.91<0.0011.151.05–1.250.002 Peripheral artery disease1.211.10–1.34<0.001​​​1.221.12–1.33<0.001 Smoking1.431.21–1.69<0.001​​​1.201.06–1.350.003Other frequent comorbidities Malignoma2.071.79–2.39<0.0010.880.78–0.990.034​​​ COPD1.261.12–1.42<0.0011.211.13–1.30<0.0011.351.23–1.47<0.001 Dementia​​​0.840.77–0.92<0.0010.690.60–0.79<0.001Obesity Grade None​​​Ref.​​Ref.​​ BMI 30–34​​​1.161.05–1.290.0031.030.90–1.180.706 BMI 35–39​​​1.211.09–1.35<0.0011.060.92–1.230.414 BMI >40​​​1.431.27–1.61<0.0011.311.13–1.52<0.001 Not spec.​​​1.121.03–1.230.0101.131.00–1.270.048Medical attention and surveillance measures Albuminuria m. (yearly)0.680.54–0.84<0.0010.820.72–0.940.0030.830.70–1.000.044 Creatinine m. (yearly)0.640.58–0.70<0.001​​​0.870.80–0.950.001 DMP diabetes​​​0.870.81–0.94<0.0010.850.77–0.940.002Pharmacologic therapy Diuretic therapy1.581.42–1.75<0.0013.893.61–4.18<0.0011.571.43–1.72<0.001 RAS Blockade0.890.81–0.980.020​​​​​​ SGLT2 Inhibition​​​​​​0.510.28–0.910.024OR, odds ratio; CI-95, 95% confidence interval; Ref., reference group; CAD, coronary artery disease; Atrial fib., atrial fibrillation; COPD, chronic obstructive pulmonary disease; m., measurement. Not spec.: not specified. Only factors with significant contribution are shown. Logistic regression analysis for subsequent death, heart failure and admission to the intensive care unit (ICU) OR, odds ratio; CI-95, 95% confidence interval; Ref., reference group; CAD, coronary artery disease; Atrial fib., atrial fibrillation; COPD, chronic obstructive pulmonary disease; m., measurement. Not spec.: not specified. Only factors with significant contribution are shown. Effect of Surveillance and Therapeutic Measures Reflected in Claims Data on Renal and Overall Outcome Risks Regarding surveillance measures, participation in checkup investigations and diabetes management programs as well as yearly albuminuria measurements is associated with a lower risk of rRT and admission for AKI. Yearly creatinine and albuminuria measurements are also associated with decreased risks of death and ICU admission (shown in Table 3). Overall, 15.3% of patients were seen by a nephrologist at least yearly (shown in Table 1). Outpatient nephrologist contact and need for rRT, but not AKI or any of the other investigated outcomes, positively associated (1.54-fold), likely due to directed referrals (Table 3). Regarding therapies, patients who received an ACE inhibitor or AT1 blocker for at least 80% of observed time experienced less AKI. Patients receiving diuretic therapy showed higher rates of rRT and AKI. SGLT2 inhibitor prescription, which at the time of data collection were licensed neither for CKD nor for heart failure, associated with a significantly lower rate of rRT (0.36-fold) and ICU admission (0.51-fold). Claims Data Associate with Pharmacologic Therapies of CKD-Related Comorbidities Baseline claims data are significantly associated with subsequent prescription of therapies directed at CKD-associated metabolic complications (all investigated pharmacologic therapies and blood transfusions; shown in Fig. 3, online suppl. Tables 2 and 3). Prescription analyses were based on ATC-coded outpatient medication claims. The addition of CKD stage significantly added to AUC for all tested conditions. Regarding individual factors, contact with a nephrologist was the most prominent factor influencing likelihood of prescription (OR = 5.23 for calcium-metabolism-related prescription, 4.33 for therapy of hyperkalemia or hyperphosphatemia, 4.61 for metabolic acidosis, and 7.43 for prescription of ESA and HIF stabilizers) (shown in online suppl. Tables 2, 3). CKDIV compared to CKDIII resulted in 2.66-fold higher calcium-metabolism-related, 1.67-fold higher hyperkalemia or hyperphosphatemia-related, 2.40-fold anti-acidemic, 2.69-fold higher erythropoietic, and 1.18-fold higher anti-hyperuricemic therapy rates. Supportive therapy prescription for CKD-related metabolic abnormalities, anemia, and gout in relation to claims data. a–e Prescription of pharmacologic therapy for calcium and vitamin D metabolism (a), hyperkalemia and hyperphosphatemia (b), metabolic acidosis (c), and erythropoietic agents (ESA and HIF stabilizers) as well as erythrocyte transfusion (d) and hyperuricemia therapies (e, codes are detailed in online suppl. Table 1) by models with and without inclusion of CKD stage. Full models are given in online suppl. Tables 2, 3. Discussion Our analysis of a large German statutory insurance company cohort demonstrates prognostic value of a combination of demographic, diagnostic, and procedural claims data for renal and overall adverse events in CKD patients in the absence of biochemical laboratory data. CKD prevalence in our 2016 cohort was 1.4% CKDIII and 0.4% CKDIV according to billing codes. These numbers are lower than CKD prevalence estimates reported from clinically characterized cohorts with biochemical CKD ascertainment from Germany and other European populations [19, 35–37], southern Denmark [38], and estimates from France [39]. However, similar discrepancies between claims-based and biochemically defined CKD prevalence have repeatedly been described in administrative database analyses and likely reflect undercoding of earlier CKD stages in routine care [19, 39, 40]. In a recent effort to relate CKD coding to actual data [40], sensitivity of claims data was 0.2 for CKDIII and 0.36 for CKDIV, which would suggest our data represent a true prevalence of 7% and 1.1%, respectively. This corresponds to the range of published proportions. In our cohort, albuminuria measurements were billed for only 5.5% of CKDIII and 6.3% of CKDIV patients. This is remarkably lower than a recent analysis of a cohort of type II diabetic CKDIV patients from the USA, of which 32% received albuminuria screening [41]. Current advances in therapy, namely, approval of SGLT2i for CKD and the associated marketing campaigns, may impact on this. An observation in the USA indeed indicates a doubling of proteinuria assessment [41], albeit with limited detectable diagnostic or therapeutic consequences. Analyzing this and cohorts recruited 2022 and later in comparison to our present data will be informative for procedures in Germany also in this regard. Co-existing conditions were common and increased with CKD stage in agreement with observations in similar settings [42]. Diabetes mellitus carried a prominent discriminatory value for all assessed clinical outcomes. This corresponds to a parallel observation in inpatients claims data from 2016–2022 assessing CKD diabetes [43] and highlights the role of diabetes in renal and cardiovascular disease. In our cohort, 3.8% of individuals coded as stage CKDIII and 13.3% of CKDIV required rRT during the observation period. In a recent large meta-analysis, this was reported for 2% CKDIII (3.3% in IIIB) and 12.4% CKDIV patients with biochemical stage classification using a combination of serum creatinine and cystatin C [1]. These published populations studied over a period from 1980 to 2011 were younger (59 ± 12 years versus 76 ± 11 years in our cohort) and followed for a longer period of time (10.8 ± 4.1 years versus 4.0 ± 2.2 years in our cohort), precluding direct comparison. Still, ranges are consistent with adequate assessment of kidney function by coded CKD severity in our cohort. Stepwise logistic regression analysis using claims data selected from prior knowledge demonstrates discriminatory value for kidney outcomes rRT and hospital admission for AKI as well as death, hospital admission for heart failure, and ICU admission. A similar prediction model of rRT from US claims data operated successfully for a 2 year period [44]. Our findings are, therefore, consistent with previous administrative and claims-based CKD analyses demonstrating that routinely collected healthcare data can provide meaningful prognostic information despite the absence of detailed biochemical characterization. Our analysis of rRT during the 6-year observation period including coded CKD stage yielded an AUC of 0.87, which is in the range of the 5 years AUC of an algorithm that chiefly employs biochemical values, resulting in an AUC of 0.85–0.89 [5]. Direct comparison between claims-based and biochemically informed prediction models such as the Kidney Failure Risk Equation (KFRE) should be interpreted cautiously as the underlying variables and intended applications differ substantially. While KFRE-based models primarily rely on laboratory parameters, the present analyses incorporated demographic characteristics, comorbidity burden, and healthcare utilization variables within a multimorbid elderly population, which may partly explain the observed discriminatory performance. CKD associates with a range of metabolic derangements. Without access to biochemical data, we employed corresponding outpatient prescriptions as surrogates for clinically relevant abnormalities. Therapies for calcium metabolism, hyperkalemia and hyperphosphatemia, metabolic acidosis, anemia, and hyperuricemia were all more commonly prescribed in CKDIV than CKDIII. Importantly, a combination of claims data significantly associated with subsequent prescriptions directed at CKD-related metabolic changes. CKD stage and nephrologist contact ranged among the most prominent factors in logistic regression for all tested, most likely also reflecting referral patterns and restricted prescription licenses e.g., for erythropoiesis stimulating agents. Active surveillance, checkup and disease management program participation associated with lowered adverse outcome risks in our cohort. Regular nephrologist contact was recorded for 10.95% of CKDIII and 29% for CKDIV patients. Rates are comparable to recent investigation from northern Germany of biochemically verified CKD, where 2% of CKDIII, 8% of CKDIIIb, and 25% of CKDIV patients were seen by a nephrologist [45], but lower than the rate of one-time nephrologist contact of 65% of CKDIV-diagnosed patients suggesting some a nationwide German claims cohort [46]. It significantly associated with adverse renal but not other outcomes. Together with increased prescription for CKD-associated therapies, this likely reflects a referral bias. Patients receiving more intensive surveillance or specific therapies may systematically differ from untreated individuals regarding comorbidity burden, disease stability, healthcare access, or physician selection. Therefore, the present findings primarily demonstrate prognostic associations within claims data rather than treatment effects. The data are somewhat controversial to German claims data 2009–2012 [47], where a benefit from nephrologist referral at documented CKDIII and earlier for renal and overall outcomes was detected. However, our study did not address earlier CKD stages. Renin-angiotensin system (RAS) inhibitor prescription rates in our cohort were similar to recently published cohorts from Sweden and the USA [48]. Continuous prescription of RAS blockers associated with lower adverse event rates in our analysis. However, these results may also reflect prescription to a clinically stable patient population, including electrolyte levels and kidney function. KDIGO recently added SGLT2 inhibition to standard CKD therapy [2]. Our cohort was collected before this and may serve as a baseline for assessing impact of SGLT2 inhibitor prescription, and also recently approved effective, but more costly, potassium binders in similar populations. Risk assessment for CKD-related outcomes requires improvement [6, 49]. The population selected for outcome analysis in our project was billed for the presence of CKDIII and IV, but only received limited laboratory monitoring. Biochemical values are major factors and appear to continue to be so in current evaluations [50–52]. Possibly, inclusion of therapeutic interventions not only as outcome measure as in our study, but also as surrogate risk factors could improve discriminatory value in the absence of measured biochemical values. This hypothesis needs to be assessed with advanced statistical and possibly machine learning methods. It should also be tested whether patients with incomplete or missing CKD codes at a high risk of CKD-related complications can be identified with large-scale claims data-derived algorithms. Administrative claims data analyses inherently carry several limitations. CKD stage assignment relied on ICD-10 coding and not on directly available biochemical measurements, making misclassification of CKD severity possible. In addition, claims-based diagnoses depend on physician coding practices and reimbursement-related documentation processes in outpatient and inpatient care, and coding quality and completeness may, therefore, vary between providers and healthcare settings. The low frequency of some surveillance-related variables, particularly albuminuria measurements, likely reflects both incomplete testing and incomplete coding within routine care. Comorbidities and outcomes may also be incompletely or inconsistently coded across healthcare settings, while prescription claims reflect dispensed medication rather than confirmed medication intake or adherence. Confounding by indication might be present, especially regarding pharmacologic therapies, specialist care, and healthcare utilization variables. At the same time, several methodological aspects strengthen the robustness and clinical relevance of the present analysis. The study is based on a large, unselected real-world population with longitudinal follow-up over 6 years. Major endpoints such as renal replacement therapy, death, and hospital admissions are closely linked to reimbursement processes and, therefore, less susceptible to underreporting. We additionally combined diagnostic, procedural, outpatient service, and prescription data using ICD, OPS, EBM, and ATC coding systems. To improve specificity of outpatient diagnoses, these were only considered present if repeatedly coded across at least 2 quarters or supported by corresponding medication prescriptions. Our cohort was derived from a single large German statutory health insurer and mainly consisted of elderly and multimorbid individuals. Therefore, the findings may not be generalizable to younger populations or healthcare systems with different structures. However, the demographic and comorbidity profile of the cohort reflects a clinically representative real-world CKD stage III–IV population frequently encountered in routine nephrology and primary care practice. In summary, CKD-related and overall outcomes significantly associated with baseline claims data in CKDIII and CKDIV patients. Advanced operationalization of these and further claims data should be investigated for improved risk assessment without the need of additional data collection. Statement of Ethics This retrospective analysis employed routinely collected data for reimbursement that can be used for scientific research without further approval or consent according to German law (Social Law V, paragraph 295). Conflict of Interest Statement The authors have no conflicts of interest to declare. Funding Sources Seb.S., V.S., T.M., and S.vV. are supported by Gemeinsamer Bundesausschuss “PRED(i)CKD-Prognostic risk evaluation with data in chronic kidney disease” (VSF_TS_2024-059). Sv.V. is member of the Excellence-Cluster “Immunosensation3” (EXC2151), which is funded as an institution by DFG (390873048). This work was supported by the Open Access Publication Fund of the University of Bonn. The funding sources had no role in the study design, collection, analysis, or interpretation of data, writing of the manuscript or the decision to submit the article for publication. Author Contributions S.S. and V.S.: conceptualization, data curation, project administration, review, and editing; O.D. and J.R.: data curation; T.M.: conceptualization and data curation; and S.vV.: conceptualization, data curation, project administration, and writing – original draft.