Abstract
Rationale
Potassium repletion is common in critically ill patients. However, practice patterns and outcomes related to different intensive care unit (ICU) potassium repletion strategies are unclear.
Objectives
1) Describe potassium repletion practices in critically ill adults; 2) compare the effectiveness of potassium repletion strategies; and 3) compare effectiveness and safety of specific potassium repletion thresholds on patient outcomes.
Methods
This was a retrospective analysis of the PINC AI Healthcare Database (2016–2022), including all critically ill adults admitted to an ICU on Hospital Day 1 and with a serum potassium concentration measured on Hospital Day 2. We determined the frequency of potassium repletion (any formulation) at each measured serum potassium concentration in each ICU, then classified ICUs as having threshold-based (a large increase in potassium repletion rates at a specific serum potassium concentration) or probabilistic (linear relationship between serum concentration and the repletion probability) patterns of repletion. Between patients in threshold-based and probabilistic repletion ICUs, we compared outcomes (primary outcome: potassium repletion frequency). We reported unadjusted percentages per exposure group and the adjusted odds ratios (from hierarchical regression models) for each outcome. Among patients in threshold-based ICUs with the most common repletion thresholds (3.5 mEq/L and 4.0 mEq/L), we conducted regression discontinuity analyses to examine the effectiveness of potassium repletion at each potassium threshold.
Results
We included 190,490 patients in 88 ICUs; 35.0% received at least one dose of potassium on the same calendar day. Rates of potassium repletion were similar between 22 threshold-based strategy ICUs (33.5%) and 22 probabilistic strategy ICUs (36.4%). There was no difference in the adjusted risk of potassium repletion between patients admitted to threshold-based strategy ICUs versus probabilistic strategy ICUs (adjusted odds ratio, 1.09; 95% confidence interval [CI], 0.76–1.57). In regression discontinuity analysis, crossing the 3.5 mEq/L threshold from high to low potassium levels resulted in a 39.1% (95% CI, 23.7–42.4) absolute increase in potassium repletion but no change in other outcomes. Similarly, crossing the 4.0 mEq/L threshold resulted in a 36.4% (95% CI, 22.4–42.2) absolute increase in potassium repletion but no change in other outcomes.
Conclusions
Potassium repletion is common in critically ill patients and occurs over a narrow range of “normal” potassium levels (3.5–4.0 mEq/L); use of a threshold-based repletion strategy to guide potassium repletion in ICU patients is not associated with clinically meaningful differences in outcomes.
Keywords: potassium, critical illness, clinical practice patterns
Potassium dysregulation occurs in almost 30% of patients with critical illness (1), and hypokalemia and hyperkalemia are associated with increased risks of cardiac dysrhythmias and death after myocardial infarction and in general ICU populations (1–4). Potassium repletion to treat or prevent hypokalemia is common in critically ill patients, although practices vary between types of intensive care units (ICUs) and individual clinicians (4–6).
Potassium may be repleted in an ad hoc manner by clinicians or via protocols that provide standing orders for potassium repletion when serum potassium levels fall below a predetermined threshold. Although these threshold-based repletion strategies are common in specific ICU populations (7, 8), practice patterns for potassium repletion and outcomes associated with ad hoc versus threshold-based potassium repletion strategies have not been investigated. Furthermore, the safety and effectiveness of specific threshold-based repletion (e.g., 3.5 mEq/L, 4.0 mEq/L) is unclear.
In this study, we sought to describe patterns of potassium repletion practices (by determining the frequency of repletion) in critically ill adults, identify ICU-level variation in potassium repletion practices and detectable potassium repletion thresholds (threshold-based or probabilistic), and, in the subgroup of patients in ICUs with threshold-based repletion strategies, evaluate the clinical effectiveness of specific serum potassium repletion thresholds to prevent hypokalemia, hyperkalemia, and related adverse outcomes.
Methods
Data Source and Cohort
We conducted a retrospective study using the PINC AI Healthcare Database, an enhanced claims-based database capturing approximately 25% of United States hospitalizations (9). PINC member hospitals submit patient-level clinical, administrative, and cost data to the database to inform quality improvement and hospital operations; deidentified data are available for purchase by independent investigators. The dataset includes International Classification of Diseases, Tenth Revision (ICD-10) diagnosis and procedure codes and daily itemized charge codes (for services, procedures, medications, and equipment). A smaller subset of hospitals in the database (approximately 15%) also contribute data captured from electronic medical records, including time- and date-stamped laboratory values (Premier Labs) that were used in this study. We included adult (age ⩾ 18 yr) patients admitted to an ICU contributing data to the Premier Labs dataset on the first day of hospital admission (identified using charge codes) between January 2016 and September 2022. We included patients with at least one measured serum potassium concentration ⩾1 and ⩽7 mEq/L (to exclude extreme values) measured on the second day of admission. We excluded potassium values on the first day of admission, because potassium repletion during the resuscitative phase of critical illness may not depend on potassium thresholds alone (e.g., diabetic ketoacidosis). For patients with more than one recorded potassium level, we selected the first potassium value. We excluded patients with a history of renal failure (using present-on-admission ICD-10 codes for comorbidities adapted from Gagne and colleagues [10, 11]) or with acute renal dysfunction (2) (for whom routine or protocol-based potassium repletion may be uncommonly used). For patients with more than one hospitalization, we selected one at random to reduce the risk of survivor bias. To increase the stability of ICU-level practice pattern estimates, we excluded patients admitted to ICUs with fewer than 1,000 potassium measurements recorded in the study dataset. Study Day 0 was defined as the second day of hospitalization (the day for which ICU repletion strategies were examined) to capture routine potassium repletion to treat or prevent mild hypokalemia during the ICU stay, not repletion related to more severe hypokalemia present on admission. In a secondary cohort, we included patients with and without renal failure.
Determination of ICU Repletion Strategies
To characterize ICU-level potassium repletion practices, we determined the frequency of potassium repletion (defined as any charge code for potassium chloride, acetate, bicarbonate, phosphorus, or gluconate via parenteral or enteral route) on the same day as potassium measurement at each measured serum potassium level (e.g., 3.5, 3.6, 3.7 mEq/L, etc.) in each ICU. A priori we sought to identify ICUs that had 1) threshold-based repletion strategies, defined as large increases in rates of potassium repletion across specific potassium concentration thresholds (e.g., 3.5 or 4.0 mEq/L); and 2), probabilistic repletion strategies, defined as small and approximately linear increases in the frequency of potassium repletion over decreasing serum potassium levels (a pattern that likely reflects ICUs with no specific protocol but in which the likelihood of potassium administration increases with lower potassium levels). The schematic in Figure E1 in the data supplement outlines the approach to identification and classification of ICU-level potassium repletion strategies in the study dataset. For each ICU, repletion strategies were defined using the following steps: 1) quantified the largest increase in potassium repletion frequency as serum potassium levels decreased between 4.5 to 3.0 mEq/L in 0.1mEq/L increments; 2) classified ICUs with a maximum increase in potassium repletion rate above the 75th percentile of all ICUs as “threshold-based repletion strategy ICUs,” ICUs below the 25th percentile as “probabilistic repletion strategy ICUs,” and all other ICUs as having an “unclear repletion strategy”; 3) among threshold-based repletion strategy ICUs, identified the serum potassium level at which the maximum increase in potassium repletion frequency occurred, then designated the potassium level 0.1 mEq/L above that value as the ICU-specific potassium repletion threshold (that is, values below this threshold were more likely to be associated with potassium administration). ICUs between the 25th and 75th percentile (i.e., those defined as having unclear repletion strategies) were excluded from subsequent outcome analyses.
Exposure and Outcomes
The exposure of interest was admission to an ICU with a threshold-based repletion strategy. Admission to an ICU with a probabilistic repletion strategy was the comparator of interest.
The primary outcome was frequency of potassium repletion on Study Day 0. Additional outcomes ascertained on Study Days 0 and 1 (Hospital Days 2 and 3) were the receipt of medications used for arrhythmia management (amiodarone, β blockers, non-dihydropyridine calcium channel blockers [diltiazem and verapamil], digoxin, or adenosine), calcium gluconate or chloride, cardioversion (ICD-10 procedure code 5A2204Z), hyperkalemia (defined as serum potassium >5.5 mEq/L excluding the index potassium level), hypokalemia (defined as serum potassium <3.0 mEq/L [excluding the index potassium concentration]), the cost of potassium repletion, and the total cost of all therapies. We also examined the next measured serum potassium level (among patients with a second potassium measurement) and the composite outcome of hospital death or discharge to hospice. Outcome definitions are described in Table E1. In a sensitivity analysis, we redefined hyperkalemia, limiting to values present on Study Day 1 to avoid inclusion of potassium levels checked during potassium repletion that might be artificially elevated.
Covariates
We included patient demographics, comorbidities (10), and severity-of-illness measures, including the presence of baseline arrhythmia, congestive heart failure, and electrolyte disorders, serum potassium level, treatments, or procedures (cardioversion, receipt of furosemide, amiodarone use, calcium gluconate or chloride, major surgery, and major cardiac surgery), and hospital-level characteristics as covariates (see Table E1 for definitions). Covariates identified from ICD-10 diagnosis codes were limited to those designated as “present on admission”; all other covariates were ascertained on Study Day −1 (Hospital Day 1).
Statistical Analysis
Covariates stratified by exposure status were summarized using counts and proportions for categorical variables and medians and interquartile ranges (IQRs) for continuous variables. Absolute standardized mean differences (SMDs) were used to compare differences in covariates between exposure groups with SMDs <0.1 suggesting that covariates were similar between groups.
Effectiveness of Potassium Repletion Strategies
We used two approaches to examine the effectiveness of potassium repletion strategies: 1) we limited the cohort to patients admitted to probabilistic or threshold-based ICUs (excluding patients admitted to ICUs with unclear strategies at risk of misclassification) and used multivariable hierarchical generalized linear models to compare outcomes among patients admitted to probabilistic versus threshold-based ICUs (with any threshold between 3.5 mEq/L and 4.0 mEq/L); and 2) we created cohorts limited to patients admitted to threshold-based ICUs with either 3.5 mEq/L or 4.0 mEq/L thresholds and used regression discontinuity to compare outcomes across repletion thresholds to identify the effectiveness of potassium repletion at specific thresholds.
Hierarchical regression models included ICU as a random effect and adjusted for all covariates to compare outcomes between patients admitted to probabilistic and threshold-based ICUs. We used logistic and linear hierarchical models for dichotomous and continuous outcomes, respectively. From these models, we reported the effect estimates (logistic models: adjusted odds ratios [aORs]; linear models: adjusted β coefficients) and 95% confidence intervals (CIs) for associations between admission to a threshold-based repletion strategy ICU versus probabilistic repletion strategy ICU and each outcome. We also reported the unadjusted percentage of patients meeting the outcome in each exposure group for dichotomous outcomes and the unadjusted mean outcome in each exposure group for continuous outcomes. Because of a higher rate of probabilistic potassium repletion strategies observed in coronary care units and cardiovascular ICUs, we considered the possibility of heterogeneity of treatment effect by ICU type by including an interaction term between exposure and ICU type (dichotomized as coronary care unit/cardiovascular ICUs vs. other ICU types) post hoc. To assess for heterogeneity of treatment effects of potassium repletion strategies on subsequent potassium levels based on the use of enteral or intravenous potassium repletion, we included an interaction term between exposure and use of enteral potassium repletion.
In regression discontinuity analyses, the exposure of interest is defined by a change in the likelihood of treatment across a range of continuous values (i.e., the first potassium level measured on Study Day 0). Because there is expected random error in serum potassium measurements (12) and because of factors exogenous to the causal structure, patients who have potassium levels slightly above or slightly below the repletion threshold are expected to have similar characteristics and are considered pseudorandomized (13). Thus, when required assumptions are met, a regression discontinuity design reduces the risk of both measured and unmeasured confounding. Given that potassium repletion at the thresholds was not deterministic (i.e., not all patients with potassium levels below the thresholds received potassium repletion, and some patients with potassium levels above the thresholds received potassium repletion), the regression discontinuity analysis mimicked the intention-to-treat effect of a randomized clinical trial that did not achieve perfect compliance with the intervention of interest (a threshold-based repletion strategy).
Regression discontinuity analyses were used to test the effectiveness of specific potassium repletion thresholds of 3.5 mEq/L and 4.0 mEq/L on patient outcomes. We quantified discontinuity at each threshold for the same outcomes as in the comparative effectiveness analysis of threshold-based versus probabilistic repletion strategies. We assessed for balance across the threshold of observed variables (falsification tests) by testing for discontinuity at the treatment threshold for age, major surgery on or before Hospital Day 1, Gagne comorbidity score present on admission, and the presence of arrhythmias on admission. To determine discontinuities at the threshold, we used local linear regression with triangular kernels and bandwidths selected using an automated algorithm that balances bias with precision (14). We reported adjusted average treatment effects as risk differences (for dichotomous outcomes) and mean differences (for continuous outcomes) with 95% CIs using robust standard errors to account for clustering of patients within ICUs.
The Boston University Institutional Review Board reviewed and approved the study protocol. Analyses were conducted using R software.
Results
Determination of ICU Repletion Strategies
We included 190,490 patients admitted to 88 distinct ICUs (Figure E2). The overall median first serum potassium level on Study Day 0 was 4.0 mEq/L (IQR, 3.7–4.4 mEq/L). A total of 66,695 (35.0%) patients received potassium repletion on Study Day 0. The overall frequency of potassium repletion across the range of serum potassium values showed potential discontinuities (i.e., large abrupt changes in use) in the frequency of potassium repletion at potassium levels of 3.5 and 4.0 mEq/L (Figure E3). We classified 22 ICUs as having threshold-based repletion strategies and 22 ICUs as having probabilistic repletion strategies; the remaining 44 were considered unclear (Figures E4–E6). Among threshold-based repletion ICUs, 9 (40.9%) had a threshold of 3.5 mEq/L, 2 (9.1%) had a threshold of 3.6 mEq/L, 1 (4.5%) had a threshold of 3.8 mEq/L, and 10 (45.5%) had a threshold of 4.0 mEq/L. The median largest difference in potassium repletion frequency across consecutive potassium levels was 42% among threshold-based repletion strategy ICUs and 19% among probabilistic repletion strategy ICUs. ICUs with identifiable threshold-based repletion strategies had abrupt changes in potassium repletion frequency across thresholds, whereas ICUs with detectable probabilistic repletion strategies demonstrated near-linear relationships between potassium levels and potassium repletion frequency (Figure 1). Characteristics of ICUs by repletion strategy are shown in Table 1.
Figure 1.
Frequency of potassium repletion by potassium concentration by intensive care unit repletion strategy.
Table 1.
Characteristics of ICUs based on detectable potassium repletion strategy
| Characteristic | Probabilistic Repletion Strategy | Threshold-based Repletion Strategy | Unclear Repletion Strategy |
|---|---|---|---|
| ICUs, n | 22 | 22 | 44 |
| ICU type | |||
| CICU/CCU/CVICU | 5 (22.7) | 3 (13.6) | 11 (25.0) |
| General/mixed | 13 (59.1) | 10 (45.5) | 30 (68.2) |
| MICU | 0 (0.0) | 4 (18.2) | 0 (0.0) |
| SICU | 3 (13.6) | 3 (13.6) | 2 (4.5) |
| Trauma | 1 (4.5) | 2 (9.1) | 1 (2.3) |
| Safety net hospital | 4 (18.2) | 12 (54.5) | 13 (29.5) |
| Urban location | 19 (86.4) | 18 (81.8) | 39 (88.6) |
| Teaching hospital | 12 (54.5) | 17 (77.3) | 26 (59.1) |
| Hospital beds | |||
| 0–99 | 1 (4.5) | 0 (0.0) | 0 (0.0) |
| 100–199 | 1 (4.5) | 2 (9.1) | 0 (0.0) |
| 200–299 | 2 (9.1) | 2 (9.1) | 10 (22.7) |
| 300–399 | 3 (13.6) | 2 (9.1) | 10 (22.7) |
| 400–499 | 2 (9.1) | 2 (9.1) | 7 (15.9) |
| ⩾500 | 13 (59.1) | 14 (63.6) | 17 (38.6) |
| U.S. census region | |||
| Midwest | 4 (18.2) | 9 (40.9) | 13 (29.5) |
| Northeast | 4 (18.2) | 1 (4.5) | 11 (25.0) |
| South | 14 (63.6) | 12 (54.5) | 20 (45.5) |
Definition of abbreviations: CCU = cardiac care unit; CICU = cardiac intensive care unit; CVICU = cardiovascular care unit; ICU = intensive care unit; MICU = medical intensive care unit; SICU = surgical intensive care unit.
Data are presented as n (%) unless otherwise specified.
Comparisons of Threshold-based versus Probabilistic Potassium Repletion Strategies
A total of 102,998 patients admitted to a probabilistic (51,038 [49.6%]) or threshold-based ICU (51,960 [50.4%]) were included in comparative effectiveness analyses. Baseline patient-level covariates that differed between potassium repletion strategy ICUs were race (SMD, 0.20), calcium gluconate or chloride use on Study Day −1 (SMD, 0.16), major surgery (SMD, 0.14), and major cardiac surgery (SMD, 0.15) (Table 2).
Table 2.
Baseline patient characteristics by potassium repletion strategy in intensive care unit of admission
| Overall (N = 102,992) | Probabilistic Repletion Strategy (n = 51,915) | Threshold-based Repletion Strategy (n = 51,083) | SMD | |
|---|---|---|---|---|
| Age, yr | 64 (52–74) | 64 (53–74) | 64 (52–74) | 0.02 |
| Sex | 0.03 | |||
| Female | 45,594 (44.3) | 22,659 (43.6) | 22,935 (44.9) | |
| Male | 57,400 (55.7) | 29,255 (56.4) | 28,145 (55.1) | |
| Unknown | 4 (0.0) | 1 (0.0) | 3 (0.0) | |
| Race | 0.20 | |||
| Asian | 1,087 (1.1) | 505 (1.0) | 582 (1.1) | |
| Black | 13,817 (13.4) | 6,700 (12.9) | 7,117 (13.9) | |
| Other | 5,638 (5.5) | 1,735 (3.3) | 3,903 (7.6) | |
| Unknown | 3,403 (3.3) | 1,615 (3.1) | 1,788 (3.5) | |
| White | 79,053 (76.8) | 41,360 (79.7) | 37,693 (73.8) | |
| Gagne Comorbidity Score (10) | 3 (3–3) | 3 (3–3) | 3 (3–3) | 0.02 |
| Individual comorbidities | ||||
| Cardiac arrhythmias | 6,053 (5.9) | 3,277 (6.3) | 2,776 (5.4) | 0.04 |
| Congestive heart failure | 6,027 (5.9) | 3,354 (6.5) | 2,673 (5.2) | 0.05 |
| Electrolyte disorders | 6,053 (5.9) | 3,277 (6.3) | 2,776 (5.4) | 0.05 |
| Sepsis present on admission (21) | 14,052 (13.6) | 6,661 (12.8) | 7,391 (14.5) | 0.01 |
| Organ dysfunctions present on admission | ||||
| Cardiovascular | 13,961 (13.6) | 7,506 (14.5) | 6,455 (12.6) | 0.05 |
| Respiratory | 30,038 (29.2) | 15,726 (30.3) | 14,312 (28.0) | 0.05 |
| Neurologic | 12,979 (12.6) | 6,281 (12.1) | 6,698 (13.1) | 0.03 |
| Hematologic | 7,164 (7.0) | 3534 (6.8) | 3,630 (7.1) | 0.01 |
| Hepatic | 896 (0.9) | 469 (0.9) | 427 (0.8) | 0.01 |
| Medications and therapies received on Study Day −1 | ||||
| Furosemide | 9,079 (8.8) | 4,981 (9.6) | 4,098 (8.0) | 0.06 |
| Amiodarone | 5,429 (5.3) | 2,929 (5.6) | 2,500 (4.9) | 0.03 |
| Calcium gluconate or chloride | 9,079 (8.8) | 4,981 (9.6) | 4,098 (8.0) | 0.16 |
| Cardioversion | 5,429 (5.3) | 2,929 (5.6) | 2,500 (4.9) | 0.03 |
| Major surgery | 53,696 (52.1) | 28,872 (55.6) | 24,824 (48.6) | 0.14 |
| Cardiac surgery | 15,580 (15.1) | 9,223 (17.8) | 6,357 (12.4) | 0.15 |
| Serum potassium concentration on Study Day −1, mEq/L* | 4.0 (3.7–4.4) | 4.1 (3.7–4.4) | 4.0 (3.7–4.4) | 0.00 |
Definition of abbreviation: SMD = standardized mean difference.
Data are presented as n (%) or median (interquartile range).
Missing in 15.9% of included patients.
Potassium repletion frequency on Study Day 0 was common in both ICU types; it was administered to 33.5% of patients in threshold-based repletion strategy ICUs and 36.4% of patients in probabilistic repletion strategy ICUs. In the hierarchical model, there was no association between admission to a threshold-based ICU (vs. probabilistic) and the receipt of potassium on Study Day 0 (aOR, 1.09; 95% CI, 0.76–1.57). There was no evidence of interaction (i.e., heterogeneity of treatment effect) between ICU repletion strategy and admission to a coronary care unit or cardiovascular ICU and potassium repletion (P value for interaction, 0.61) or between ICU repletion strategy and the use of enteral potassium repletion (P value for interaction, 0.26). Hyperkalemia was less frequent among patients in threshold-based ICUs than those in probabilistic ICUs (2.4% vs. 3.2%; aOR, 0.60; 95% CI, 0.46–0.79); this effect was attenuated when limiting to potassium values on Study Day 1 (aOR, 0.80; 95% CI, 0.59–1.07). There were no differences in other outcomes between groups (Table 3).
Table 3.
Patient outcomes associated with admission to a threshold-based repletion ICU
| Outcome | Percentage of Cohort Patients Admitted to Probabilistic Repletion Strategy ICUs | Percentage of Cohort Patients Admitted to Threshold-based Repletion Strategy ICUs | Adjusted Odds Ratio (95% CI) for Threshold-based Repletion Strategy (Ref. Probabilistic Repletion Strategy) |
|---|---|---|---|
| Potassium repletion on Study Day 0 | 36.4 | 33.5 | 1.09 (0.76 to 1.57) |
| Medications for arrhythmia use on Study Day 0 or 1 | 32.6 | 37.7 | 0.96 (0.77 to 1.20) |
| Calcium gluconate or chloride use on Study Day 0 or 1 | 17.7 | 15.5 | 0.93 (0.53 to 1.65) |
| Cardioversion on Study Day 0 or 1 | 0.5 | 0.4 | 0.80 (0.51 to 1.25) |
| Hyperkalemia on Study Day 0 or 1 | 3.2 | 2.4 | 0.60 (0.46 to 0.79) |
| Hypokalemia on Study Day 0 or 1 | 3.2 | 3.1 | 0.85 (0.65 to 1.10) |
| Hospital death or discharge to hospice on Study Day 0 or 1 | 13.9 | 13.2 | 1.02 (0.86 to 1.22) |
| Mean among Patients Admitted to Probabilistic Repletion Strategy ICUs | Mean among Patients Admitted to Threshold-based Repletion Strategy ICUs | Adjusted β Coefficients (95% CI) for Threshold-based repletion strategy (Ref. Probabilistic repletion strategy) | |
|---|---|---|---|
| Hospital costs on Study Day 0 or 1, U.S. dollars | 28,844 | 24,770 | 3,019 (−6,986 to 12,934) |
| Potassium repletion costs on Study Day 0 or 1, U.S. dollars | 49 | 42 | 6 (−13 to 24) |
| Subsequent serum potassium concentration, mEq/L | 4.07 | 4.05 | −0.1 (−0.2 to 0.01) |
Definition of abbreviations: CI = confidence interval; ICU = intensive care unit.
Effectiveness of Potassium Repletion at 3.5 mEq/L and 4.0 mEq/L
The regression discontinuity analysis included 28,046 patients admitted to 9 ICUs with thresholds of 3.5 mEq/L and 16,166 patients admitted to 10 ICUs with thresholds of 4.0 mEq/L. ICUs with 3.5 mEq/L thresholds administered at least one dose of potassium to 27% of cohort patients; those with 4.0 mEq/L thresholds administered potassium to 46%. Among ICUs with 3.5 mEq/L thresholds, crossing the threshold from higher to lower potassium resulted in a 39.1 (95% CI, 23.7–42.4) absolute percentage increase in the rate of potassium repletion (Figure 2) but no changes in other study outcomes measured on Day 0 or 1, including receipt of medications for arrhythmia, calcium receipt, cardioversion, hyperkalemia, hypokalemia, in-hospital death or discharge to hospice, potassium repletion, and total hospital costs, or subsequent potassium levels (Table E2). Similarly, in ICUs with 4.0 mEq/L thresholds, crossing the threshold from higher to lower potassium resulted in a 36.4 (95% CI, 22.4–42.2) absolute percentage increase in potassium repletion rate but no changes in other outcomes. There was no evidence of discontinuity at either threshold in the falsification tests.
Figure 2.
Discontinuity at thresholds for potassium repletion and use of medications for arrhythmia. (A and B) Intensive care units (ICUs) with a 3.5 mEq/L threshold. (C and D) ICUs with a 4.0 mEq/L threshold. (A and C) Illustrate potassium repletion on Study Day 0. (B and D) Illustrate use of medications for arrhythmia on Study Days 0 and 1.
Secondary Cohort Including Patients with and without Renal Failure
A total of 304,304 patients (116 ICUs) were included in the secondary cohort that did not exclude patients with renal failure from analysis. We classified 29 ICUs as having threshold-based, 29 as probabilistic, and 58 as unclear repletion strategies. As in the primary cohort, there was evidence of discontinuities (i.e., large abrupt changes in use) in threshold-based repletion at potassium levels of 3.5 mEq/L and 4.0 mEq/L (Figure E7). Similar to the primary cohort analysis, there was no association between admission to a threshold-based ICU (vs. probabilistic) and the receipt of potassium on Study Day 0 (aOR, 0.78; 95% CI, 0.59–1.02). Odds of hyperkalemia were lower in patients admitted to threshold-based ICUs (aOR, 0.68; 95% CI, 0.53–0.86), but other secondary outcomes were similar between ICU types (Table E3).
Discussion
In this large multicenter clinical dataset–based study of ICU potassium repletion practices and patient outcomes, we identified serum potassium threshold-based potassium repletion strategies and distinguished these strategies from other repletion approaches. Except for a lower rate of hyperkalemia among patients who received potassium repletion in ICUs with threshold-based practices, we identified few differences in clinical or cost outcomes between ICUs that administered potassium according to potassium threshold values versus ICUs using other repletion strategies. Among patients admitted to ICUs with detectable repletion thresholds, we found higher rates of potassium receipt in ICUs with a higher threshold concentration (4.0 mEq/L vs. 3.5 mEq/L) but no differences in clinical outcomes between patients with potassium levels just above or below the most used thresholds. Results were similar when including patients with renal failure. Our results suggest that potassium repletion in U.S. ICUs occurs over a relatively narrow range of “normal” potassium measurements and that, within this range, different strategies to replete potassium are not strongly associated with clinical outcomes.
Our results should be considered in the context of previous studies. Randomized trials of potassium repletion in pediatric and adult patients after cardiac surgery have not tested whether potassium should be repleted. Instead, these trials, predicated on the assumption that potassium repletion was necessary, compared the efficacy of enteral versus parenteral potassium administration to maintain identical serum potassium levels (15) or examined clinical outcomes after randomization to serum goals of 4.0 mEq/L versus 4.5 mEq/L (16). Observational work examining electrolyte repletion in large clinical datasets has described inconsistent clinical practice and use of personal heuristics (such as replacing potassium when other electrolytes are replaced or for patients with conditions associated with hypokalemia) that were associated with substantial variation in the degree and route of potassium repletion administered (5, 6).
Although this study focuses on potassium administration at the ICU level (and could not identify administration by individual prescribers), a majority of ICUs in the study cohort did not have discernable patterns of potassium repletion attributable to specific serum potassium levels, precluding those ICUs and patients from analyses. This, and the small number of ICUs included in specific threshold analyses, potentially limits the generalizability of study findings. As in prior studies, potassium repletion patterns in unclassifiable ICUs may be attributable to differences in individual practices, differences in patient characteristics, or low adherence to local protocols (if available). Because the most common thresholds for potassium replacement were 3.5 and 4.0 mEq/L, decisions by clinicians or protocols may reflect digit preference bias, in which decisions to intervene are more likely when terminal values equal 0 or 5 (17).
Strengths and Limitations
Our study has strengths. The large number of included ICUs and heterogeneous critically ill patients increases the generalizability of the results. By using ICU-level repletion strategy as the exposure of interest, we decreased the risk of indication bias (18). Use of regression discontinuity to compare potassium repletion to no repletion at thresholds of 3.5 and 4.0 mEq/L minimizes the risk of both measured but unaddressed and unmeasured confounding. Although future randomized trials of different repletion strategies may be useful, baseline differences in local practice potentially threaten trial enrollment and protocol adherence and should be addressed intentionally.
Our study also has limitations. First, because data were deidentified, we could not ascertain reasons for observed differences in potassium repletion patterns between units (such as local practice, protocol availability, or protocol adherence). Second, the study examined potassium repletion over a relatively narrow range of serum potassium levels generally considered normal. Thus, the results should not be used to guide repletion strategies at more extreme potassium levels, where the likelihood of observable benefit or risk is likely to be higher (4). Moreover, we did not examine doses of potassium administered; undetectable differences in recommended potassium doses for a given serum potassium concentration between ICU types may explain the lower rate of hyperkalemia on the day of potassium repletion observed in threshold-based ICUs. Similarly, results of the regression discontinuity analyses that do not suggest clinically meaningful differences in outcomes between repletion and no repletion may be attributed to administration of very small doses of potassium near threshold levels that were unlikely to substantially change serum potassium levels or clinical outcomes. The finding of lower odds of hyperkalemia in patients admitted to threshold-based ICUs was attenuated when limiting to potassium levels on Study Day 1. These results could suggest that 1) potassium repletion strategies are not associated with risks of hyperkalemia 24 hours after repletion; or 2) the association between hyperkalemia and repletion strategies when including potassium levels on Study Day 0 could have included potassium levels drawn during repletion, leading to spuriously high values. Some measured clinical outcomes, such as receipt of antiarrhythmic medications, cannot distinguish between continuation of home medications and initiation of a new medication during hospitalization. Last, we did not examine the full range of outcomes previously attributed to electrolyte repletion protocols, including variability in serum potassium values, shorter times between abnormal laboratory value results and electrolyte repletion, and higher satisfaction of nurses and physicians (4, 8, 19, 20).
Conclusions
In a large U.S. cohort of critically ill patients, we found wide variation in the application of potassium repletion strategies across ICUs. However, we found no evidence that potassium repletion guided by a potassium concentration threshold heuristic were associated with improved outcomes compared with alternative strategies and found no evidence that repletion at common potassium concentration thresholds of 3.5 or 4.0 mEq/L was associated with patient outcomes. These results suggest that over the relatively narrow range of “normal” potassium at which critically ill patients typically receive routine potassium repletion, the approach used by individual ICUs is not a discernable driver of outcomes.
Footnotes
Supported by National Institutes of Health National Center for Advancing Translational Sciences grants 1KL2TR001411 and 1UL1TR001430 and the Boston University Chobanian & Avedisian School of Medicine Department of Medicine Career Investment Award.
Author Contributions: N.A.B. and E.A.V. take responsibility for the integrity of the work as a whole, from inception to published article. All authors substantially contributed to the conception and design of this study. N.A.B. acquired the data. All authors were involved in the interpretation of data. N.A.B. and E.A.V. drafted the manuscript, and all authors revised it critically for important intellectual content. All authors read and approved the final manuscript.
This article has a data supplement, which is accessible from this issue’s table of contents at www.atsjournals.org.
Author disclosures are available with the text of this article at www.atsjournals.org.
References
- 1. Weir MR, Espaillat R. Clinical perspectives on the rationale for potassium supplementation. Postgrad Med . 2015;127:539–548. doi: 10.1080/00325481.2015.1045814. [DOI] [PubMed] [Google Scholar]
- 2. Goyal A, Spertus JA, Gosch K, Venkitachalam L, Jones PG, Van den Berghe G, et al. Serum potassium levels and mortality in acute myocardial infarction. JAMA . 2012;307:157–164. doi: 10.1001/jama.2011.1967. [DOI] [PubMed] [Google Scholar]
- 3. Bouadma L, Mankikian S, Darmon M, Argaud L, Vinclair C, Siami S, et al. OUTCOMEREA Study Group Influence of dyskalemia at admission and early dyskalemia correction on survival and cardiac events of critically ill patients. Crit Care . 2019;23:415. doi: 10.1186/s13054-019-2679-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Engelhardt LJ, Balzer F, Müller MC, Grunow JJ, Spies CD, Christopher KB, et al. Association between potassium concentrations, variability and supplementation, and in-hospital mortality in ICU patients: a retrospective analysis. Ann Intensive Care . 2019;9:100. doi: 10.1186/s13613-019-0573-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Joseph TT, DiMeglio M, Huffenberger A, Laudanski K. Behavioural patterns of electrolyte repletion in intensive care units: lessons from a large electronic dataset. Sci Rep . 2018;8:11915. doi: 10.1038/s41598-018-30444-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Ghannam M, Malihi P, Laudanski K. Examination of electrolyte replacements in the ICU utilizing MIMIC-III dataset demonstrates redundant replacement patterns. Healthcare (Basel) . 2021;9:1373. doi: 10.3390/healthcare9101373. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Couture J, Létourneau A, Dubuc A, Williamson D. Evaluation of an electrolyte repletion protocol for cardiac surgery intensive care patients. Can J Hosp Pharm . 2013;66:96–103. doi: 10.4212/cjhp.v66i2.1231. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Kanji Z, Jung K. Evaluation of an electrolyte replacement protocol in an adult intensive care unit: a retrospective before and after analysis. Intensive Crit Care Nurs . 2009;25:181–189. doi: 10.1016/j.iccn.2009.03.004. [DOI] [PubMed] [Google Scholar]
- 9.PINC AI Applied Sciences. 2023. https://offers.pinc-ai.com/PINC-AI-Healthcare-Database-White-Paper-LP.html
- 10. Gagne JJ, Glynn RJ, Avorn J, Levin R, Schneeweiss S. A combined comorbidity score predicted mortality in elderly patients better than existing scores. J Clin Epidemiol . 2011;64:749–759. doi: 10.1016/j.jclinepi.2010.10.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Sun JW, Rogers JR, Her Q, Welch EC, Panozzo CA, Toh S, et al. Adaptation and validation of the combined comorbidity score for ICD-10-CM. Med Care . 2017;55:1046–1051. doi: 10.1097/MLR.0000000000000824. [DOI] [PubMed] [Google Scholar]
- 12. Asirvatham JR, Moses V, Bjornson L. Errors in potassium measurement: a laboratory perspective for the clinician. N Am J Med Sci . 2013;5:255–259. doi: 10.4103/1947-2714.110426. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Bor J, Moscoe E, Mutevedzi P, Newell ML, Bärnighausen T. Regression discontinuity designs in epidemiology: causal inference without randomized trials. Epidemiology . 2014;25:729–737. doi: 10.1097/EDE.0000000000000138. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Calonico S, Cattaneo MD, Farrell MH. On the effect of bias estimation on coverage accuracy in nonparametric inference. J Am Stat Assoc . 2018;113:767–779. [Google Scholar]
- 15. Rehman Siddiqu NU, Merchant Q, Hasan BS, Rizvi A, Amanullah M, Rehmat A, et al. Comparison of enteral versus intravenous potassium supplementation in hypokalaemia in paediatric patients in intensive care post cardiac surgery: open-label randomised equivalence trial (EIPS) BMJ Open . 2017;7:e011179. doi: 10.1136/bmjopen-2016-011179. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Hoekstra M, Hessels L, Rienstra M, Yeh L, Lansink AO, Vogelzang M, et al. Computer-guided normal-low versus normal-high potassium control after cardiac surgery: no impact on atrial fibrillation or atrial flutter. Am Heart J . 2016;172:45–52. doi: 10.1016/j.ahj.2015.10.020. [DOI] [PubMed] [Google Scholar]
- 17. Thavarajah S, White WB, Mansoor GA. Terminal digit bias in a specialty hypertension faculty practice. J Hum Hypertens . 2003;17:819–822. doi: 10.1038/sj.jhh.1001625. [DOI] [PubMed] [Google Scholar]
- 18. Johnston SC. Combining ecological and individual variables to reduce confounding by indication: case study: subarachnoid hemorrhage treatment. J Clin Epidemiol . 2000;53:1236–1241. doi: 10.1016/s0895-4356(00)00251-1. [DOI] [PubMed] [Google Scholar]
- 19. Pearson DJ, Sharma A, Lospinoso JA, Morris MJ, McCann ET. Effects of electrolyte replacement protocol implementation in a medical intensive care unit. J Intensive Care Med . 2018;33:574–581. doi: 10.1177/0885066616679593. [DOI] [PubMed] [Google Scholar]
- 20. Hijazi M, Al-Ansari M. Protocol-driven vs. physician-driven electrolyte replacement in adult critically ill patients. Ann Saudi Med . 2005;25:105–110. doi: 10.5144/0256-4947.2005.105. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Angus DC, Linde-Zwirble WT, Lidicker J, Clermont G, Carcillo J, Pinsky MR. Epidemiology of severe sepsis in the United States: analysis of incidence, outcome, and associated costs of care. Crit Care Med . 2001;29:1303–1310. doi: 10.1097/00003246-200107000-00002. [DOI] [PubMed] [Google Scholar]