Key Points
Tools for diagnosis or prediction of hyperkalemia in CKD are limited, especially in patients not using RAASi.
This study constructed a convenient nomogram and online calculator to identify the risk of hyperkalemia within 6 months in patients with advanced CKD.
Patients with CKD who are identified as high risk of hyperkalemia by the tool may benefit from intensive monitoring and early triage.
Keywords: chronic kidney disease, hyperkalemia, nomogram, risk factors
Visual Abstract
Abstract
Background
Hyperkalemia is a common and life-threatening complication of CKD. We aimed to develop and validate a nomogram that could identify the risk of hyperkalemia (≥5.5 mmol/L) in patients with CKD.
Methods
A retrospective cohort study was performed in adult patients with predialysis advanced CKD (stages ≥3) in 2020–2021 for the outcome of hyperkalemia within 6 months. The training set was used to identify risk factors of hyperkalemia. Then a nomogram was developed by multivariable logistic regression analysis. C-statistics, calibration curves, and decision curve analysis (DCA) were used, and the model was validated in the internal and two external validation sets.
Results
In total, 847 patients with advanced CKD were included. In 6 months, 28% of patients had hyperkalemia (234 out of 847). Independent risk factors were: age ≥75 years, higher CKD stages, previous event of serum potassium ≥5.0 mmol/L within 3 months, and comorbidities with heart failure, diabetes, or metabolic acidosis. Then the nomogram on the basis of the risk factors adding the use of renin-angiotensin-aldosterone system inhibitors was constructed. The C-statistic of the model was 0.76 (95% CI, 0.70 to 0.78), and was stable in both the internal validation set (0.73; 95% CI, 0.63 to 0.82) and external validation sets (0.88; 95% CI, 0.84 to 0.95 and 0.82; 95% CI, 0.72 to 0.92). Calibration curves and DCA analysis both found good performances of the nomogram.
Conclusion
A feasible nomogram and online calculator were developed and validated to evaluate the risk of hyperkalemia within 6 months in patients with advanced CKD. Patients with CKD and a high risk of hyperkalemia may benefit from intensive monitoring and early triage.
Introduction
CKD, defined as abnormalities of kidney structure or function presenting for >3 months, is a global public health problem with a prevalence of 10%–15% worldwide (1). CKD is an independent risk factor for death, cardiovascular disease, and a broad range of complications including hyperkalemia. Hyperkalemia affects approximately 20% of patients with CKD, and the prevalence is significantly higher than that of the general population (2,3). The prevalence of hyperkalemia in patients with advanced CKD was about 20%, 40%, and 50% at stages 3, 4, and 5, respectively (4).
Hyperkalemia can lead to heart palpitations, chest pain, shortness of breath, nausea, or vomiting (2). Moreover, severe hyperkalemia (serum potassium level of ≥6 mmol/L) is a life-threatening condition that requires immediate medical intervention. CKD is the main driver of the increase of developing hyperkalemia. Meanwhile, the presence of hyperkalemia in patients with CKD is associated with a worse renal prognosis, arrhythmia, and mortality (5–10). Moreover, nearly one third of patients with CKD could have repeated episodes of hyperkalemia (11). Furthermore, it was costlier to treat CKD when hyperkalemia was present. In a study of patients with CKD, those with hyperkalemia cost US$25,156 more annually than those without (12). In addition, hyperkalemia expenditures were 22% and 38% higher for Medicare and commercial patients, respectively, in CKD stages 4–5 (13). Guidelines suggest measuring serum potassium before the first prescription or up-titration of renin-angiotensin-aldosterone system inhibitors (RAASi), and in the subsequent 2 weeks (14). However, many patients with CKD lack electrolyte monitoring after RAASi initiation in current clinical settings, and patients with CKD not using RAASi may miss more detection of hyperkalemia (15). Consequently, it is important to monitor serum potassium continuously, and in a timely manner, in patients with CKD with a high risk of hyperkalemia.
The clinical risk and potential economic value highlight the importance of predicting which patients with CKD are at high risk of developing hyperkalemia (16). Traditional predictive factors associated with hyperkalemia in CKD include high CKD stages, dehydration, consuming excessive dietary potassium, heart failure (HF), diabetes, metabolic acidosis, and using potassium-increasing medications such as RAASis, potassium-sparing diuretics, and so on (9,17–21). However, convenient tools for diagnosis or prediction the risk of hyperkalemia in CKD are few (2). There is no feasible model for identifying hyperkalemia risk in patients with CKD. Therefore, we developed and validated a nomogram and online tool to identify hyperkalemia risk in patients with advanced CKD.
Methods
Study Population
This retrospective cohort study was performed according to the TRIPOD statement (22). The TRIPOD checklist is listed in Supplemental Table 1. The analysis of anonymous data was reviewed and approved by the ethical committee of participating hospitals. The informed consent of patients was waived because patients were assigned to training and validation sets by using their medical records. In total, this study included 847 eligible patients from June 2020 to June 2021. Among the 847 patients from Changzheng Hospital (n=434) and Zhabei Central Hospital (n=413), 675 (80%) were randomly assigned to the training set, whereas the other 172 were allocated to the internal validation set. Patients from the Naval Medical Center of People's Liberation Army (PLA) (n=218), and PLA 902 Hospital (n=81) were used as the external validation set 1 and external validation set 2, respectively.
Inclusion criteria: inpatients or outpatients, age ≥18 years, clinically diagnosed with CKD stage ≥3 (eGFR <60 ml/min per m2) (14), and with the status of hyperkalemia ≤6 months after enrollment. CKD was defined at enrollment by the diagnosis code. If patients tested potassium levels multiple times, the latest results were used. Any previous event of serum potassium ≥5.0 mmol/L within 3 months before enrollment was recorded. Exclusion criteria were dialysis, pregnancy, comorbidities of tumors, infectious diseases, AKI, and no relevant data. For multiple visits to the same patient, the last one that met the inclusion and exclusion criteria was selected. The staging of HF included the New York Heart Association (NYHA) functional classes (II–IV) (23).
Data Collection and Choice of Risk Factors
Hyperkalemia was defined as serum potassium ≥5.5 mmol/L. Hyperkalemia stratification (serum potassium level ≥ or <5.5 mmol/L) was used as the primary outcome. The demographic, age, sex, laboratory examination indicators, comorbidities, and drug use history were collected from the electronic medical record system from included hospitals. We searched databases (PubMed and EMBASE) using “chronic kidney disease or CKD,” “hyperkalemia,” and “risk factors” as the search terms to find published risk factors of hyperkalemia in CKD. Moreover, we consulted an expert group composed of senior nephrologists and solicited opinions on the items to be included in the model. Finally, we chose the potential risk factors as follows: age, sex, CKD stages; medical histories: cardiovascular disease, HF, previous event of blood potassium levels (PEBP) ≥5.0 mmol/L within 3 months, diabetes, metabolic acidosis, and medications: potassium supplements, RAASi (including angiotensin-converting enzyme inhibitors, angiotensin-receptor blockers, mineralocorticoid receptor antagonists, and direct renin inhibitors), potassium-sparing diuretics, other diuretics, β-blockers, nonsteroidal anti-inflammatory drugs [NSAIDs], calcineurin inhibitors, electrocardiograph, and Chinese herbal medicine. Electronic health data with a measurement of serum potassium, and serum creatinine (Scr) was extracted together with other clinical data. CKD stages were calculated by the CKD Epidemiology Collaboration formula using Scr (24).
Model Construction and Performance Evaluation
A logistic regression algorithm was then used to construct the risk-assessment model using variables in the training set. The prior compute required sample size was calculated by G power software v3.1.9.2. Logistic regression of the binary outcome variable (Y) on a binary independent variable (X) with a sample size of 797 observations (of which 80% are in group X=0 and 20% are in the group X=1) achieves 95% power at a 0.05 significance level to detect a change. This change corresponds to an odds ratio (OR) of 2.0. The primary outcome was the probability of hyperkalemia in patients with CKD at 6 months. The confounders included in the multivariate regression were determined according to the results of single-variable analysis (P<0.10) and the recommendations of clinical experts, and the conditional likelihood ratio method was used to construct the final multivariate model. Odds ratios and 95% confidence intervals (95% CIs) were reported. Akaike's information criterion and Bayesian information criterion were used to compare models. Reclassification improvement was quantified by the integrated discrimination improvement and net reclassification improvement statistics. Then, a logistic regression formula was used to calculate the risk scores for patients with CKD and the risks of hyperkalemia. Furthermore, we used the coefficients of the multivariable logistic regression model to generate a nomogram. The performance of the nomogram was assessed by the discrimination and calibration along with the Hosmer–Lemeshow test. Receiver operating characteristic curve (ROC) was used to determine the best cutoff value and evaluate the diagnostic value. Cutoff scores were used to separate patients with CKD into high- and low-risk groups. The area under ROC was calculated by concordance statistics (C-statistics) to evaluate the discrimination ability, and more than 0.7 was considered a good fit. The calibration ability was evaluated by the calibration curve which was assessed graphically by plotting the observed probabilities against the model predicted probabilities. Sensitivity analysis was performed by model comparisons. ROC analyses of the nomogram scores compared with other risk factors in the model were performed.
Model Validation
The internal validation set and two external validation sets were used to validate the performance of the model. Risk scores were calculated for each patient of the validation sets on the basis of the model constructed in the training set. Then, the C-statistic and calibration curves of the model were also assessed. Decision curve analysis was performed to determine the net benefits at different threshold probabilities by using the nomogram.
Statistical Analysis
Analyses in this study were completed using SPSS v22.0, R 4.2.0, and STATA v16. Continuous data were described by the number of patients, mean, and standard deviation, or median and quartiles by the distribution type. The t test or Mann–Whitney U test was used for comparison between groups if the data distribution was abnormal. The dichotomous variable was described by frequency and percentage; the chi-squared test or Fisher's exact test was used for the comparisons. If missing values were >20% in a variable, the variable would be deleted. All statistical tests were two sided, and a P value <0.05 was considered statistically significant.
Ethical Approval
Our study complied with the Declaration of Helsinki and was approved by the Ethics Committee of Changzheng Hospital (reference 2020SL012) and by other participating hospitals: Zhabei Central Hospital (JAZB035), Naval Medical Center of PLA (N0011), and PLA 902 Hospital (21B27).
Informed Consent
The informed consent of patients was waived using anonymous data by the Ethics Committee of Changzheng Hospital (reference 2020SL012), for patients were assigned to training and validation sets by using their medical records.
Results
Patient Characteristics
In total, 847 eligible patients who were predialysis with CKD were included. There were 769 outpatients (91%) and 78 inpatients (9%). The flow diagram was shown in Figure 1. The characteristics of the patients were summarized in Table 1. The median age was 58 (interquartile range, 47–68) years old. The incidence of hyperkalemia was 28% (234 out of 847) of patients with CKD at 6 months. CKD stages included 3a (29%), 3b (39%), 4 (15%), and 5 (17%), respectively. The characteristics of two external validation sets are listed in Supplemental Tables 2 and 3. The incidence of hyperkalemia was 29% (193 out of 675), 24% (41 out of 172), 14% (30 out of 218), and 24% (19 out of 81) in the training set, internal validation set, and external validation sets 1 and 2, respectively. By initial group comparisons, hyperkalemia was associated with older age, higher Scr, lower HCO3−, higher CKD stages, serum potassium ≥5 mmol/L, HF, more comorbidities with diabetes, metabolic acidosis, and use of β-blockers, NSAIDs, and Chinese herbal medicine (P<0.05, Table 1).
Figure 1.
The flowchart depicts the design of this study.
Table 1.
The baseline characteristics of patients with advanced CKD
| Characteristics | Total (n=847) | Nonhyperkalemia in 6 Mo (n=613) | Hyperkalemia in 6 Mo (n=234) | P |
|---|---|---|---|---|
| Age, yr | 58 (47–68) | 58 (45–68) | 60 (50–70) | 0.04a |
| <75 | 734 (87%) | 546 (89%) | 188 (81%) | 0.002a |
| ≥75 | 108 (13%) | 65 (11%) | 43 (19%) | |
| Male | 555 (66%) | 407 (66%) | 148 (63%) | 0.34 |
| Serum creatinine, µmol/L | 169.5 (131–280) | 158 (129–227) | 219 (143–448) | <0.001a |
| Serum potassium, mmol/L | 4.39 (3.87–5.07) | 4.1 (3.71–4.46) | 5.38 (5.2–5.6) | <0.001a |
| Serum HCO3−, mmol/L | 22.72±4.07 | 23.34±3.82 | 21.28±4.29 | <0.001a |
| CKD stage | <0.001a | |||
| G3a | 245 (29%) | 209 (34%) | 36 (15%) | |
| G3b | 333 (39%) | 245 (40%) | 88 (38%) | |
| G4 | 127 (15%) | 88 (14%) | 39 (17%) | |
| G5 | 142 (17%) | 71 (11%) | 71 (30%) | |
| PEBP ≥5 mmol/L | 81 (10%) | 26 (4%) | 55 (24%) | <0.001a |
| CVD | 422 (50%) | 292 (47%) | 130 (56%) | 0.10 |
| Heart failure | 59 (7%) | 32 (5%) | 27 (12%) | 0.02a |
| Diabetes | 275 (32%) | 175 (29%) | 100 (43%) | 0.001a |
| Metabolic acidosis | 132 (16%) | 65 (11%) | 67 (29%) | <0.001a |
| Drug use | ||||
| RAASi | 316 (37%) | 232 (38%) | 84 (36%) | 0.63 |
| β-blocker | 211 (25%) | 138 (23%) | 73 (31%) | 0.009a |
| Potassium supplements | 51 (6%) | 40 (7%) | 11 (5%) | 0.33 |
| Potassium-sparing diuretics | 65 (8%) | 42 (7%) | 23 (10%) | 0.15 |
| Other diuretics | 226 (27%) | 135 (22%) | 91 (39%) | <0.001a |
| NSAIDs | 55 (7%) | 33 (5%) | 22 (9%) | 0.03a |
| Calcineurin inhibitors | 46 (5%) | 36 (6%) | 10 (4%) | 0.35 |
| Chinese herbal medicine | 475 (56%) | 332 (54%) | 143 (61%) | 0.04a |
Characteristics are summarized as median (IQR), mean (SD), or frequency (%). PEBP, previous event of blood potassium levels; CVD, cardiovascular disease; RAASi, renin-angiotensin-aldosterone system inhibitors; NSAIDs, nonsteroidal anti-inflammatory drugs; IQR, interquartile range.
Significant P values.
Construction of Risk Model
In the training set, 15 factors were included in the univariable analysis (Table 2). Nine factors including age, CKD stage, HF, PEBP ≥5.0 mmol/L, diabetes, metabolic acidosis, use of β-blockers, NSAIDs, and Chinese herbal medicine were further included in the multivariate analysis (Table 2). The results showed that six factors including age ≥75 years, CKD stage, HF, PEBP ≥5 mmol/L, diabetes, and acidosis were independent risk factors of hyperkalemia in CKD. Moreover, according to the suggestions of nephrologists, model 2 adding the use of RAASi, and model 3 adding synergic potassium-increasing drugs (including RAASi, β-blockers, NSAIDs, and Chinese herbal medicine) were constructed. Comparisons of the three models found that model 2 had the smallest value of Akaike's information criterion and Bayesian information criterion. Meanwhile, the net reclassification improvement of model 2 was 0.04±0.02 (P=0.004), and the integrated discrimination improvement was 0.003±0.0007 (P<0.001), indicating that model 2 was better (Table 3).
Table 2.
Factors associated with hyperkalemia within 6 months in patients with CKD by univariable and multivariable analyses
| Univariate Logistic Regression | Multivariate Model 1 | Model 2 | Model 3 | |||||
|---|---|---|---|---|---|---|---|---|
| Variables | OR (95% CI) | P | OR (95% CI) | P | OR (95% CI) | P | OR (95% CI) | P |
| Age (≥75 yr versus <75 yr) | 1.92 (1.26 to 2.92) | 0.002a | 1.82 (1.08 to 3.04) | 0.02 | 1.80 (1.07 to 3.03) | 0.03 | 1.82 (1.09 to 3.05) | 0.02 |
| Sex (female versus male) | 1.15 (0.84 to 1.57) | 0.39 | NA | NA | NA | |||
| CKD stage | ||||||||
| G3a | Reference | Reference | Reference | Reference | ||||
| G3b | 2.09 (1.36 to 3.20) | <0.001a | 1.98 (1.19 to 3.29) | 0.008a | 2.02 (1.21 to 3.36) | 0.007a | 1.97 (1.19 to 3.28) | 0.009a |
| G4 | 2.57 (1.53 to 4.32) | <0.001a | 2.07 (1.13 to 3.79) | 0.02a | 2.08 (1.13 to 3.81) | 0.02a | 1.98 (1.08 to 3.65) | 0.03a |
| G5 | 5.81 (3.58 to 9.41) | <0.001a | 3.67 (2.04 to 6.60) | <0.001a | 3.61 (2.00 to 6.52) | <0.001a | 3.55 (1.97 to 6.40) | <0.001a |
| Cardiovascular disease | 1.29 (0.95 to 1.74) | 0.10 | NA | NA | NA | |||
| Heart failure | 1.85 (1.11 to 3.05) | 0.02a | 2.18 (1.12 to 4.26) | 0.02a | 2.21 (1.13 to 4.31) | 0.02a | 2.20 (1.13 to 4.29) | 0.02a |
| PEBP ≥5 mmol/L | 7.35 (4.80 to 11.25) | <0.001a | 4.03 (2.39 to 6.82) | <0.001a | 4.31 (2.53 to 7.34) | <0.001a | 4.03 (2.39 to 6.82) | <0.001a |
| Diabetes | 1.68 (1.23 to 2.28) | 0.001a | 1.46 (0.99 to 2.14) | 0.05a | 1.49 (1.01 to 2.19) | 0.04a | 1.42 (0.96 to 2.08) | 0.08a |
| Metabolic acidosis | 3.62 (2.44 to 5.38) | <0.001a | 1.68 (1.04 to 2.71) | 0.034a | 1.79 (1.10 to 2.90) | 0.02a | 1.67 (1.03 to 2.69) | 0.04a |
| Potassium-increasing drugs | ||||||||
| Potassium supplements | 0.71 (0.36 to 1.42) | 0.33 | NA | NA | NA | |||
| RAASi | 1.02 (0.60 to 1.73) | 0.93 | NA | 1.36 (0.76 to 2.43) | 0.30 | NA | ||
| Potassium-sparing diuretics | 1.48 (0.87 to 2.52) | 0.15 | NA | NA | NA | |||
| β-blocker | 1.56 (1.11 to 2.18) | 0.01a | NA | NA | NA | |||
| NSAIDs | 1.82 (1.04 to 3.20) | 0.036a | NA | NA | NA | |||
| Calcineurin inhibitors | 0.72 (0.347 to 1.46) | 0.353 | NA | NA | NA | |||
| Chinese herbal medicine | 1.40 (1.02 to 1.92) | 0.036a | NA | NA | NA | |||
| RAASi+β-blocker+ NSAIDs+CHM | 1.57 (1.09 to 2.28) | 0.02a | NA | NA | 1.30 (0.86 to 1.95) | 0.22 | ||
| Constant | NA | 0.11 (0.07 to 0.18) | 0.10 (0.07 to 0.17) | 0.10 (0.06 to 0.16) | ||||
Model 1 adjusted variables: age ≥75 years, CKD stage, HF, previous blood potassium ≥5 mmol/L, diabetes, and metabolic acidosis; Model 2 added RAASi on the base of Model 1; Model 3 added RAASi+β-blocker+NSAIDs+CHM on the base of Model 1. OR, odds ratio; 95% CI, 95% confidence interval; NA, not available; PEBP, previous event of blood potassium levels; RAASi, renin-angiotensin-aldosterone system inhibitors; NSAIDs, nonsteroidal anti-inflammatory drugs; CHM, Chinese herbal medicine.
P values <0.10.
Table 3.
Comparisons of the multivariate models
| Model Indicators | Model 1 | Model 2 | Model 3 |
|---|---|---|---|
| AIC | 708.44 | 703.58 | 708.89 |
| BIC | 748.99 | 748.61 | 753.95 |
| NRI | Ref | 0.04±0.02a,b | 0.03±0.03b,c |
| IDI | Ref | 0.003±0.0007b,d | 0.00002±0.0005b,e |
Model 1 was the reference model of IDI and NRI. Model 1 adjusted variables: age ≥75 years, CKD stage, HF, previous blood potassium ≥5 mmol/L, diabetes, and metabolic acidosis; Model 2 added RAASi on the base of Model 1; Model 3 added RAASi+β-blocker+NSAIDs+Chinese herbal medicine on the base of Model 1. AIC, Akaike's information criterion; BIC, Bayesian information criterion; NRI, net reclassification improvement; IDI, integrated discrimination improvement; Ref, reference; HF, heart failure; NSAIDs, nonsteroidal anti-inflammatory drugs; RAASi, renin-angiotensin-aldosterone system inhibitors.
P=0.004.
Significant results.
P=0.01.
P=0.0001.
P=0.97.
The regression formula of model 2 with seven risk factors was as follows: risk score=0.589×(age ≥75 years)+ 0.701×CKD 3b stage+0.730×CKD 4 stage+1.285×CKD 5 stage+0.792×HF+1.461×PEBP ≥5.0 mmol/L+0.397×diabetes+0.580×metabolic acidosis+0.309×RAASi−2.26. The indicator of each factor was equal to 1 if the statement was positive and was equal to 0 otherwise. The predicted probability of hyperkalemia can be calculated using 1/(1+exp [−risk score]). To provide a more convenient tool, we developed a nomogram and an online calculator (https://www.shczyy.com/mzg/page1.html) (Figure 2). Nomogram scores were calculated as follows: Score=4.0×(1 if age ≥75 years)+4.8×(1 if CKD 3b stage)+5.0×(1 if CKD 4 stage)+8.8 (1 if CKD 5 stage)+5.4×(1 if HF)+10×(1 if PEBP ≥5.0 mmol/L)+2.7×(1 if diabetes)+4.0×(1 if metabolic acidosis)+2.1×(1 if using RAASi).
Figure 2.
Characteristics in the nomogram to predict the probability of hyperkalemia within 6 months in patients with advanced CKD. Patient prognostic values are located on the axis of each variable; a line is then drawn upwards at a 90° angle to determine the number of points for the variable. Score=4.0×(1 if age ≥75 years)+4.8×(1 if CKD 3b stage)+5.0×(1 if CKD 4 stage)+8.8×(1 if CKD 5 stage)+5.4×(1 if heart failure)+10×(1 if previous event of blood potassium levels ≥5.0 mmol/L)+2.7×(1 if diabetes)+4.0×(1 if metabolic acidosis)+2.1×(1 if using RAASi). The sum of these numbers is located on the total score axis, and a line is drawn at a 90° angle downward to the hyperkalemia risk axis to determine the likelihood of hyperkalemia. Alternatively, hyperkalemia risk can be ascertained from the online calculator. The indicator of the binary factor was equal to 1 if the statement was positive and was equal to 0 otherwise. Prob, probability.
In the training set, the C-statistic of this model was 0.76 (95% CI, 0.70 to 0.78; Figure 3A). The optimal cutoff value for the nomogram score was 8 according to the Youden index. The calibration curve showed fine calibration of the model (Figure 4A). In addition, the Hosmer–Lemeshow test yielded a nonsignificant statistic (P=0.87), which indicated a good fit. The C-statistic of nomogram score was better than each of the other risk factors alone (DeLong test, P all <0.001, Figure 5), including age (C-statistic 0.55, 95% CI, 0.52 to 0.58), CKD stages (0.66; 95% CI, 0.61 to 0.70), HF (0.53; 95% CI, 0.51 to 0.56), PEBP ≥5.0 mmol/L (0.62; 95% CI, 0.59 to 0.66), diabetes (0.56; 95% CI, 0.52 to 0.60), and metabolic acidosis (0.59; 95% CI, 0.55 to 0.62).
Figure 3.
ROC curves of the nomogram. (A) Training set. (B) Internal validation set. (C) External validation set 1. (D) External validation set 2. ROC, receiver operator characteristic.
Figure 4.
Calibration curves for the nomogram. The calibration method was used to illustrate the association between actual hyperkalemia and predicted hyperkalemia. Calibration plots show the predictive, and ideal (100% agreement) curves with bootstrapping samples. The nomogram-predicted probability of hyperkalemia was plotted on the x-axis; the observed probability of hyperkalemia was plotted on the y-axis. (A) Training set. (B) Internal validation set. (C) External validation set 1. (D) External validation set 2.
Figure 5.
Comparison of ROC curves for hyperkalemia between nomogram scores and other variables. Comparison of ROC curves for hyperkalemia showing area under the curve including nomogram scores (C-statistic, 0.74; 95% confidence interval [95% CI], 0.70 to 0.78, CKD stages; 0.66; 95% CI, 0.61 to 0.70), history of heart failure (0.53; 95% CI, 0.51 to 0.56), blood potassium levels >5.0 mmol/L history (0.62; 95% CI, 0.59 to 0.66), diabetes (0.56; 95% CI, 0.52 to 0.60), and acidosis history (0.59; 95% CI, 0.55 to 0.62).
Model Validation
The discrimination performance of the nomogram was then validated in the internal validation set (C-statistic, 0.73; 95% CI, 0.63 to 0.82), which remained stable. The performance was also confirmed in the external validation set 1 and 2 with C-statistics of 0.88 (95% CI, 0.84 to 0.95) and 0.82 (95% CI, 0.72 to 0.92), respectively (Figure 3). Furthermore, the calibration plots overlapped with the ideal line in both the training set and external validation sets, showing adequate agreement of the predictive possibilities with actual observations (Figure 4).
Clinical Usefulness of the Nomogram
In all sets, decision curve analysis curves indicated that using the nomogram could add more net benefit than either the “treat all” or “treat none” strategies scheme for a wide range of threshold probability, indicating well clinical usefulness (Figure 6). Threshold probabilities for the net benefit associated with the use of this nomogram in detecting hyperkalemia ranged from 0.10 to 0.85 in the training set, from 0.10 to 0.73 in the internal validation set, and from 0.10 to 0.85 in external validation set 1 and 2.
Figure 6.
Decision curve analyses for the nomogram. The decision curve analysis graph was drawn by plotting threshold probability on the x-axis and net benefit on the y-axis, illustrating the trade-offs between benefit (true positives) and harm (false positives). The amount of net benefit varies as the threshold probability (preference or not for treatment) was varied. The dotted line represented the nomogram. The black line represented the hypothesis that all patients had hyperkalemia. The gray line represented the hypothesis that no patients had hyperkalemia. (A) Training set. (B) Internal validation set. (C) External validation set 1. (D) External validation set 2. Threshold probabilities for the net benefit associated with the use of this nomogram in detecting hyperkalemia ranged from 0.10 to 0.84 in the training set, from 0.10 to 0.73 in the internal validation set, and from 0.10 to 0.85 in external validation set 1 and 2.
Discussion
Hyperkalemia is a common and life-threatening complication in patients with CKD. Patients with CKD will spend approximately 9% of the disease time with hyperkalemia, on average (25). Although potassium binders for hyperkalemia treatment are effective and available, the early warning of hyperkalemia is still challenging, because patients are often asymptomatic, and blood potassium monitoring is underperformed (25). In this study, a nomogram and online calculator were developed and validated for the identification of hyperkalemia within 6 months in patients with advanced CKD. The nomogram had a good discriminatory ability (C-statistic =0.76) and was stable across diverse patients with advanced CKD. For patients with CKD with a clinical indication to evaluate hyperkalemia risk, the application of the nomogram or online tool may help.
To our knowledge, this is the first nomogram to evaluate the risks of hyperkalemia in advanced CKD. Compared with the real-world research of Kashihara et al. (3), the prevalence of hyperkalemia in each stage of CKD of our study was close, which indicated our data could reflect the real-world situation. Compared with traditional methods, our nomogram showed more convenience. Bandak et al. (15) followed up 69,000 patients (CKD and non-CKD) for 1 year to predict hyperkalemia after initiating RAASi. The screened-out risk factors were similar to our model. However, the calculation of the SCREAM model was complex and only applicable to patients with CKD using RAASi (15). Johnson et al. (26) developed a model to predict the quintile risk of hyperkalemia in 5171 patients with advanced CKD within 90 days of starting lisinopril. The observed risk was 0.7% and 7% for the lowest and highest quintiles, but the characteristics of model performance were not reported (26). Recently, Ajay et al. (27) developed and validated a model to identify patients with CKD at elevated risk for developing hyperkalemia over a year, using claims from a large US health care payer without external validation. There were 21 independent predictors included in the final model that made the calculation complicated (27). Compared with the studies above, this nomogram focused on the practicability and convenience of identifying the risk of hyperkalemia in patients with advanced CKD.
Hyperkalemia in CKD is the result of a comprehensive effect of multiple mechanisms, such as excessive potassium intake or production, reduced renal excretion, and imbalanced distribution (28). Risk factors in this model could contribute to hyperkalemia by different mechanisms. Advanced age is always accompanied by decreases in plasma renin activity and plasma aldosterone levels and frequent use of NSAIDs in the elderly, and is prone to hyperkalemia (29). Under normal conditions, 90% of potassium ions are excreted by the kidneys, and excess potassium ions are hard to accumulate (30). When the renal structure and function are impaired in CKD, the ability to excrete potassium decreases gradually. Although it can be partly compensated by increasing colonic excretion of potassium, it is not sufficient (31). Moreover, the risk of hyperkalemia will increase when CKD combines with comorbidities such as HF, diabetes, and so on (25). HF can lead to a reduction in renal perfusion and is always accompanied by usage of RAASis or mineralocorticoid receptor antagonists, which further increase the risk of hyperkalemia. A large study found the 5-year hyperkalemia prevalence reached up to 48% in patients with CKD and HF (27). Insulin deficiency and hypertonicity caused by hyperglycemia in diabetes can contribute to an inability to disperse high potassium load into intracellular space. In addition, metabolic acidosis, as one of the common complications of CKD, could shift potassium from the intracellular to the extracellular space and increase the serum level of potassium (30). Last but not the least, RAASis play important roles in delaying CKD progression and are the most widely used medications in CKD. However, RAASis could block the renin-angiotensin system and cause lower serum aldosterone, which impairs potassium excretion (32). RAASi medications were not significantly higher in the hyperkalemia group than the nonhyperkalemia group in this study. This may be explained by the restricted use in patients with advanced CKD in our center, because RAASi may increase the level of Scr (33,34). Although potassium binders may permit patients with advanced CKD at risk to benefit from RAASi use, whether potassium binders would affect the protective effect of RAASi in CKD was not yet clear.
This study had several limitations. First, this study used a retrospective design. The follow-up time was short, and the logistic model made the onset time of hyperkalemia not well shown. Second, this nomogram was not suitable for patients with CKD with eGFR ≥60 ml/min per m2, for it would increase the false-positive rates of hyperkalemia. Third, the sample size of this study was moderate. Inpatients and outpatients were included together, which might lead to selection bias. Future prospective designs could be used to optimize the model and increase the sample size of the study. It may even learn from the practices by establishing a CKD registry, then a large sample of data can be used.
In conclusion, this study constructed a nomogram and online calculator to identify the risk of hyperkalemia within 6 months in patients with advanced CKD. The nomogram to noninvasively identify the high risk of hyperkalemia using personal data may benefit patients with CKD with a high risk from intensive monitoring and early triage.
Disclosures
All authors have nothing to disclose.
Funding
This work was supported by the Shanghai Science and Technology Innovation Plan (22Y11905500), National Natural Science Foundation of China (82070705, 81770670, and 81873595), Shanghai Municipal Key Clinical Specialty (shslczdzk02503), fundings from naval medical center (21M3201, 21M3202), Shanghai Jing'an District Health Commission Research Project (2022MS04), and the Research Projects of Shanghai Science and Technology Committee (17411972100).
Acknowledgments
The authors would like to acknowledge the support of Dr. Daoji Xue for his help.
Author Contributions
Z. Mao and C. Mei conceptualized the study; X. Luo, C. Mei, C. Xue, X. Ye, and C. Zhou were responsible for the data curation; X. Hu, X. Luo, C. Xue, X. Ye, and C. Zhou were responsible for the formal analysis; Z. Mao and C. Mei were responsible for the funding acquisition; J. Chen, X. Hu, Z. Mao, C. Mei, C. Xue, J. Xu, and B. Yang were responsible for the investigation; Y. Lu, X. Luo, C. Xue, and X. Ye were responsible for the methodology; Y. Lu was responsible for the project administration; J. Chen, X. Hu, Z. Mao, B. Yang, and L. Zhang were responsible for the resources; J. Chen, Y. Lu, Z. Mao, and L. Zhang were responsible for the software; Z. Mao and L. Zhang provided supervision; L. Zhang was responsible for the validation; X. Luo and L. Zhang were responsible for the visualization; C. Xue wrote the original draft; Z. Mao, C. Mei, C. Xue, J. Xu, and B. Yang reviewed and edited the manuscript; all authors read and approved the final manuscript.
Data Sharing Statement
The anonymized datasets generated and/or analyzed during this study are available from the corresponding author on reasonable request. Raw Data/Source Data, Figshare, https://figshare.com/s/998e76b82a77e8c548f2.
Supplemental Material
This article contains the following supplemental material online at http://kidney360.asnjournals.org/lookup/suppl/doi:10.34067/KID.0004752022/-/DCSupplemental
TRIPOD checklist: Prediction model development. Download Supplemental Table 1, PDF file, 398 KB (397.5KB, pdf) .
Baseline characteristics of patients with CKD in external validation set 1. Download Supplemental Table 2, PDF file, 398 KB (397.5KB, pdf) .
Baseline characteristics of patients with CKD in external validation set 2. Download Supplemental Table 3, PDF file, 398 KB (397.5KB, pdf) .
References
- 1.Zhang L, Wang F, Wang L, Wang W, Liu B, Liu J, Chen M, He Q, Liao Y, Yu X, Chen N, Zhang JE, Hu Z, Liu F, Hong D, Ma L, Liu H, Zhou X, Chen J, Pan L, Chen W, Wang W, Li X, Wang H: Prevalence of chronic kidney disease in China: A cross-sectional survey. Lancet 379: 815–822, 2012. 10.1016/S0140-6736(12)60033-6 [DOI] [PubMed] [Google Scholar]
- 2.Clase CM, Carrero JJ, Ellison DH, Grams ME, Hemmelgarn BR, Jardine MJ, Kovesdy CP, Kline GA, Lindner G, Obrador GT, Palmer BF, Cheung M, Wheeler DC, Winkelmayer WC, Pecoits-Filho R; Conference Participants : Potassium homeostasis and management of dyskalemia in kidney diseases: Conclusions from a Kidney Disease: Improving Global Outcomes (KDIGO) Controversies Conference. Kidney Int 97: 42–61, 2020. 10.1016/j.kint.2019.09.018 [DOI] [PubMed] [Google Scholar]
- 3.Kashihara N, Kohsaka S, Kanda E, Okami S, Yajima T: Hyperkalemia in real-world patients under continuous medical care in Japan. Kidney Int Rep 4: 1248–1260, 2019. 10.1016/j.ekir.2019.05.018 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Moranne O, Froissart M, Rossert J, Gauci C, Boffa JJ, Haymann JP, M’rad MB, Jacquot C, Houillier P, Stengel B, Fouqueray B; NephroTest Study Group : Timing of onset of CKD-related metabolic complications. J Am Soc Nephrol 20: 164–171, 2009. 10.1681/ASN.2008020159 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Kovesdy CP, Regidor DL, Mehrotra R, Jing J, McAllister CJ, Greenland S, Kopple JD, Kalantar-Zadeh K: Serum and dialysate potassium concentrations and survival in hemodialysis patients. Clin J Am Soc Nephrol 2: 999–1007, 2007. 10.2215/CJN.04451206 [DOI] [PubMed] [Google Scholar]
- 6.Collins AJ, Pitt B, Reaven N, Funk S, McGaughey K, Wilson D, Bushinsky DA: Association of serum potassium with all-cause mortality in patients with and without heart failure, chronic kidney disease, and/or diabetes. Am J Nephrol 46: 213–221, 2017. 10.1159/000479802 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Silva-Cardoso J, Brito D, Frazão JM, Ferreira A, Bettencourt P, Branco P, Fonseca C: Management of RAASi-associated hyperkalemia in patients with cardiovascular disease. Heart Fail Rev 26: 891–896, 2021. 10.1007/s10741-020-10069-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.An JN, Lee JP, Jeon HJ, Kim DH, Oh YK, Kim YS, Lim CS: Severe hyperkalemia requiring hospitalization: Predictors of mortality. Crit Care 16: R225, 2012. 10.1186/cc11872 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Jain N, Kotla S, Little BB, Weideman RA, Brilakis ES, Reilly RF, Banerjee S: Predictors of hyperkalemia and death in patients with cardiac and renal disease. Am J Cardiol 109: 1510–1513, 2012. 10.1016/j.amjcard.2012.01.367 [DOI] [PubMed] [Google Scholar]
- 10.Fu W, Zhang A, Ma L, Jia L, Chhetri JK, Chan P: Severity of frailty as a significant predictor of mortality for hemodialysis patients: A prospective study in China. Int J Med Sci 18: 3309–3317, 2021. 10.7150/ijms.51569 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Nilsson E, Gasparini A, Ärnlöv J, Xu H, Henriksson KM, Coresh J, Grams ME, Carrero JJ: Incidence and determinants of hyperkalemia and hypokalemia in a large healthcare system. Int J Cardiol 245: 277–284, 2017. 10.1016/j.ijcard.2017.07.035 [DOI] [PubMed] [Google Scholar]
- 12.Betts KA, Woolley JM, Mu F, Xiang C, Tang W, Wu EQ: The cost of hyperkalemia in the United States. Kidney Int Rep 3: 385–393, 2017. 10.1016/j.ekir.2017.11.003 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Golestaneh L, Alvarez PJ, Reaven NL, Funk SE, McGaughey KJ, Romero A, Brenner MS, Onuigbo M: All-cause costs increase exponentially with increased chronic kidney disease stage. Am J Manag Care 23[Suppl]: S163–S172, 2017 [PubMed] [Google Scholar]
- 14.Levin A, Stevens PE: Summary of KDIGO 2012 CKD Guideline: Behind the scenes, need for guidance, and a framework for moving forward. Kidney Int 85: 49–61, 2014. 10.1038/ki.2013.444 [DOI] [PubMed] [Google Scholar]
- 15.Bandak G, Sang Y, Gasparini A, Chang AR, Ballew SH, Evans M, Arnlov J, Lund LH, Inker LA, Coresh J, Carrero JJ, Grams ME: Hyperkalemia after initiating renin-angiotensin system blockade: The Stockholm Creatinine Measurements (SCREAM) Project. J Am Heart Assoc 6: e005428, 2017. 10.1161/JAHA.116.005428 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Vinson AJ, Zanjir W, Nallbani M, Goldstein J, Swain J, Clark DA, More KM, Manderville JR, Fok PT, Wiemer H, Tennankore KK: Predictors of hyperkalemia among patients on maintenance hemodialysis transported to the emergency department by ambulance. Kidney360 3: 615–626, 2022. 10.34067/KID.0008132021 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Lazich I, Bakris GL: Prediction and management of hyperkalemia across the spectrum of chronic kidney disease. Semin Nephrol 34: 333–339, 2014. 10.1016/j.semnephrol.2014.04.008 [DOI] [PubMed] [Google Scholar]
- 18.Thomsen RW, Nicolaisen SK, Hasvold P, Garcia-Sanchez R, Pedersen L, Adelborg K, Egfjord M, Egstrup K, Sørensen HT: Elevated potassium levels in patients with congestive heart failure: Occurrence, risk factors, and clinical outcomes: A Danish population-based cohort study. J Am Heart Assoc 7: e008912, 2018. 10.1161/JAHA.118.008912 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Adelborg K, Nicolaisen SK, Hasvold P, Palaka E, Pedersen L, Thomsen RW: Predictors for repeated hyperkalemia and potassium trajectories in high-risk patients - A population-based cohort study. PLoS One 14: e0218739, 2019. 10.1371/journal.pone.0218739 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Gwoo S, Kim YN, Shin HS, Jung YS, Rim H: Predictors of hyperkalemia risk after hypertension control with aldosterone blockade according to the presence or absence of chronic kidney disease. Nephron Clin Pract 128: 381–386, 2014. 10.1159/000369138 [DOI] [PubMed] [Google Scholar]
- 21.Korgaonkar S, Tilea A, Gillespie BW, Kiser M, Eisele G, Finkelstein F, Kotanko P, Pitt B, Saran R: Serum potassium and outcomes in CKD: Insights from the RRI-CKD cohort study. Clin J Am Soc Nephrol 5: 762–769, 2010. 10.2215/CJN.05850809 [DOI] [PMC free article] [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. Ann Intern Med 162: 55–63, 2015. 10.7326/M14-0697 [DOI] [PubMed] [Google Scholar]
- 23.Raphael C, Briscoe C, Davies J, Ian Whinnett Z, Manisty C, Sutton R, Mayet J, Francis DP: Limitations of the New York Heart Association functional classification system and self-reported walking distances in chronic heart failure. Heart 93: 476–482, 2007. 10.1136/hrt.2006.089656 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Inker LA, Eneanya ND, Coresh J, Tighiouart H, Wang D, Sang Y, Crews DC, Doria A, Estrella MM, Froissart M, Grams ME, Greene T, Grubb A, Gudnason V, Gutiérrez OM, Kalil R, Karger AB, Mauer M, Navis G, Nelson RG, Poggio ED, Rodby R, Rossing P, Rule AD, Selvin E, Seegmiller JC, Shlipak MG, Torres VE, Yang W, Ballew SH, Couture SJ, Powe NR, Levey AS; Chronic Kidney Disease Epidemiology Collaboration : New creatinine- and cystatin c-based equations to estimate GFR without race. N Engl J Med 385: 1737–1749, 2021. 10.1056/NEJMoa2102953 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Kovesdy CP: Management of hyperkalaemia in chronic kidney disease. Nat Rev Nephrol 10: 653–662, 2014. 10.1038/nrneph.2014.168 [DOI] [PubMed] [Google Scholar]
- 26.Johnson ES, Weinstein JR, Thorp ML, Platt RW, Petrik AF, Yang X, Anderson S, Smith DH: Predicting the risk of hyperkalemia in patients with chronic kidney disease starting lisinopril. Pharmacoepidemiol Drug Saf 19: 266–272, 2010. 10.1002/pds.1923 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Sharma A, Alvarez PJ, Woods SD, Dai D: A model to predict risk of hyperkalemia in patients with chronic kidney disease using a large administrative claims database. Clinicoecon Outcomes Res 12: 657–667, 2020. 10.2147/CEOR.S267063 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Lehnhardt A, Kemper MJ: Pathogenesis, diagnosis and management of hyperkalemia. Pediatr Nephrol 26: 377–384, 2011. 10.1007/s00467-010-1699-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Turgut F, Balogun RA, Abdel-Rahman EM: Renin-angiotensin-aldosterone system blockade effects on the kidney in the elderly: Benefits and limitations. Clin J Am Soc Nephrol 5: 1330–1339, 2010. 10.2215/CJN.08611209 [DOI] [PubMed] [Google Scholar]
- 30.Harris AN, Grimm PR, Lee HW, Delpire E, Fang L, Verlander JW, Welling PA, Weiner ID: Mechanism of hyperkalemia-induced metabolic acidosis. J Am Soc Nephrol 29: 1411–1425, 2018. 10.1681/ASN.2017111163 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Hunter RW, Bailey MA: Hyperkalemia: Pathophysiology, risk factors and consequences. Nephrol Dial Transplant 34[Suppl 3]: iii2–iii11, 2019. 10.1093/ndt/gfz206 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Mercier K, Smith H, Biederman J: Renin-angiotensin-aldosterone system inhibition: overview of the therapeutic use of angiotensin-converting enzyme inhibitors, angiotensin receptor blockers, mineralocorticoid receptor antagonists, and direct renin inhibitors. Prim Care 41: 765–778, 2014. 10.1016/j.pop.2014.08.002 [DOI] [PubMed] [Google Scholar]
- 33.Jun M, Jardine MJ, Perkovic V, Pilard Q, Billot L, Rodgers A, Rogers K, Gallagher M: Hyperkalemia and renin-angiotensin aldosterone system inhibitor therapy in chronic kidney disease: A general practice-based, observational study. PLoS One 14: e0213192, 2019. 10.1371/journal.pone.0213192 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Hasegawa T, Nishiwaki H, Ota E, Levack WM, Noma H: Aldosterone antagonists for people with chronic kidney disease requiring dialysis. Cochrane Database Syst Rev 2: CD013109, 2021. 10.1002/14651858.CD013109.pub2 [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.
Supplementary Materials
TRIPOD checklist: Prediction model development. Download Supplemental Table 1, PDF file, 398 KB (397.5KB, pdf) .
Baseline characteristics of patients with CKD in external validation set 1. Download Supplemental Table 2, PDF file, 398 KB (397.5KB, pdf) .
Baseline characteristics of patients with CKD in external validation set 2. Download Supplemental Table 3, PDF file, 398 KB (397.5KB, pdf) .