Inpatient hospitalisation and mortality rate trends from 2004 to 2014 in the USA: a propensity score-matched case-control study of hyperkalaemia


Unmatched analysis

To achieve our objective regarding prevalence, we required the use of an unmatched dataset. There was a total of 24 941 608 discharge records of patients aged ≥18 years in the NIS from 2004 to 2014 with presence of CHF, CKD/ESRD, AKI or T2DM, which represent a total of 120 513 483 (±2 312 391) inpatient discharges in the USA. In this cohort, we found a total of 1 397 573 records containing hyperkalaemia, which represent a total of 6 761 577 (±149 409) discharges in the USA. This corresponds to an average annual prevalence of 5.61%, which increased over time from 4.94%±0.07% in 2004 to 6.37%±0.04% in 2014, a relative increase of 28.9% (p<0.0001, figure 1). Partly due to the large sample size, significant differences between groups were observed in every variable examined (table 1); however, the distributions of age, gender, HF and hospital characteristics were similar between those who did versus did not have hyperkalaemia. African Americans and Hispanics had a higher risk of hyperkalaemia than Caucasians. Hospitalisations including hyperkalaemia had higher rates of renal dysfunction (acute and chronic) and major/extreme loss of function (APR-DRG severity).

Figure 1
Figure 1

Prevalence of hyperkalaemia in inpatient hospitalisations including congestive heart failure, chronic kidney disease (and end-stage renal disease), acute kidney injury and/or type II diabetes mellitus.

Table 1

Patient characteristics of the unmatched and matched cohorts according to hyperkalaemia presence

Inpatient mortality rates were significantly higher for cases with versus without hyperkalaemia (average absolute difference=4.0%, average relative difference=97.81%, p<0.0001), and the rate decreased non-uniformly between groups over time, decreasing at a faster rate for cases with hyperkalaemia (10.91%±0.17% to 6.23%±0.08%) than for cases without hyperkalaemia (4.81%±0.05% to 3.8%±0.03%) (pyear<0.0001, pinteraction<0.0001, figure 2).

Figure 2
Figure 2

Annual in-hospital mortality rates (with SE bars) for the unmatched cohort according to the presence of hyperkalaemia in hospitalisations including congestive heart failure, chronic kidney disease (and end-stage renal disease), acute kidney injury and/or type II diabetes mellitus.

Matched analysis

To achieve our objective regarding inpatient mortality rates while accounting for confounders, we performed PS matching. After matching, we had a total of 2 606 462 records, representing 12 517 269 (±174 562) hospital discharges. The unweighted records reflect the 1:1 matching (ie, 1 303 231 records in each group), but they represent an odd number of discharges due to records having unequal weights. Patient characteristics were well balanced, with standardised differences all <0.10 (table 1). Note that because we excluded cases of hyperkalaemia as the primary diagnosis for the matched analyses, the cases with hyperkalaemia and their characteristics are not identical to those in the unmatched cohort.

Inpatient mortality rates were significantly higher for cases with versus without hyperkalaemia (average absolute difference=1.71%, average relative difference=25.3%, p<0.0001), and the rate decreased uniformly between groups over time, decreasing from 11.49%±0.17% to 6.43%±0.08% for cases with hyperkalaemia and from 9.67%±0.13% to 5.05%±0.07% for cases without hyperkalaemia (p<0.0001, figure 3).

Figure 3
Figure 3

Annual in-hospital mortality rates (with SE bars) for the propensity score-matched cohort according to the presence of hyperkalaemia in hospitalisations including congestive heart failure, chronic kidney disease (and end-stage renal disease), acute kidney injury and/or type II diabetes mellitus.


In this study, considering adult inpatient hospitalisations with HF, CKD/ESRD, AKI and/or T2DM, we found a relative increase of 28.9% in hyperkalaemia prevalence (from 4.94% in 2004 to 6.37% in 2014). We found that hospitalisations in which hyperkalaemia occurred were far more likely to be severe in nature. Accordingly, we found that the presence of hyperkalaemia was associated with a higher rate of inpatient mortality. Further, after controlling for primary diagnosis, severity of illness, comorbidities, hospital characteristics and sociodemographics, we found that the presence of hyperkalaemia continued to play a significant role in inpatient mortality risk. We also observed significant reductions in inpatient mortality over time.

Our work reiterates and extends findings from Betts and colleagues, who determined that the prevalence of hyperkalaemia among patients with CKD and/or HF increased from 4.95% to 6.35% (a relative increase of 28.2%) using insurance claims records and laboratory test results from 2010 to 2014 in the Truven MarketScan claims and encounters database.17 The nearly 30% increase in hyperkalaemia prevalence in Betts’ study, as well as in our current examination of inpatient hospitalisations may be partially explained by the ageing population, increasing comorbidity burden and need for chronic/multiple medications.3 4 Additionally, our timeframe is large enough such that improved abilities and/or standards of documentation may have been adopted by hospitals over time.18 For example, it is possible that the implementation of specialised tools within electronic health systems over time may have made the documentation of multiple diagnoses easier.19 Similarly, another possible explanation is that general awareness of hyperkalaemia may have increased over time and that physicians became more likely to screen for it. For example, searching PubMed for the term ‘hyperkalemia’ yields 206 and 357 papers for 2004 and 2014, respectively.

Our findings extend those of Singer and colleagues’ cross-sectional study which determined that hyperkalaemia was independently associated with greater risk of inpatient admission (80% vs 39% from patients in the emergency department with moderate hyperkalaemia vs normal potassium levels, respectively) and mortality (5.5% vs 0.8% among those with moderate hyperkalaemia vs normal potassium levels, respectively).20 Similarly, Davis and colleagues found that having severe hyperkalaemia increased the risk of inpatient mortality by 58.5% compared with having mild hyperkalaemia (19.5% vs 12.3%).21 Cheungpasitporn and colleagues found mild hyperkalaemia to carry an associated 22% increased risk of inpatient mortality among those with CKD, after adjusting for confounders.22 While we do not know the severity of hyperkalaemia in our study, our results are similar in that the presence of hyperkalaemia was associated with an average 25% increase in the risk for mortality in the matched analysis and a 98% increase in the unmatched analysis. In general, hyperkalaemia’s association with increased risk of mortality may simply be reflective of a more severe overall presentation, or it may contribute to death by complicating an already difficult-to-treat disease state, or even more directly by inducing life-threatening cardiac arrhythmias.1 23 Our observation of mortality rates declining over time may be reflective of the large percentage of records with CKD in this study, as it has been shown that CKD mortality rates in Medicare beneficiaries have declined over time but remain significantly higher than the rates observed in patients without CKD.24 Further, the declining rates may be partially attributable to advancements in technology and medical care delivery, including medications. For example, increased use of point-of-care potassium testing could have resulted in faster delivery of care.25

Although we observed a significant increase in its prevalence, as well as a higher mortality rate for those who have it, preventing and treating hyperkalaemia is possible. In some cases, particularly patients with CKD at risk for chronic hyperkalaemia, a potassium-restricted diet may be beneficial.26 For cases of drug-induced hyperkalaemia, interrupting the prescription may be a solution; however, new challenges may arise if the medication was for the management of a chronic condition, which is often true.2 Alternatively, diuretics may be used to increase potassium excretion via urine and dialysis may be used to remove excess potassium from blood. In the setting of a hyperkalaemic emergency, an intravenous infusion of calcium and insulin may be used to both protect the heart and cause a cellular shift of potassium. Another treatment for hyperkalaemia is potassium-binding medication, which expels excess potassium through faecal matter.27 One such drug is sodium polystyrene sulfonate (SPS), which has been used since the late 1950s, but is associated with serious gastrointestinal side effects (and even colonic necrosis in rare cases) and has a relatively low adherence rate.28 Two additional drugs, sodium zirconium cyclosilicate and patiromer, help patients achieve and maintain normal potassium levels.29 These have advantages over SPS in that they are associated with fewer side effects and they may be efficacious regardless of renin-angiotensin-aldosterone system inhibitors (RAASi) and/or diet.25 These newer drugs received Food and Drug Administration approval after our study timeframe, so they do not explain our observed reduction in mortality rate; however, it is of interest to determine whether these rates have further declined since their availability. For patients taking medication for chronic diseases, incorporating a pharmacist into a team-based management approach may help protect against hyperkalaemia.30

The study was designed to examine any record with HF, CKD/ESRD, AKI or T2DM. Doing so provided a very large and rich dataset for studying hyperkalaemia trends in inpatient hospitalisations. Due to the broad inclusion criteria of these analyses, this work did not shed light on disease-specific inferences. It is possible that the trends observed in this overall cohort may not hold for each specific disease group. In this paper, we overcame the inherent imbalance of characteristics between hospitalisations with versus without hyperkalaemia by performing additional analyses on a PS-matched dataset, which made our conclusions more robust. Further, we conducted the PS matching within specific primary diagnoses because it is our intention to perform subgroup analyses according to primary diagnosis in future work.

Limitations of this study include that the timeframe under evaluation ended in 2014; this was due to availability of data and to maintain consistency with ICD-9-CM coding. We acknowledge that there may be additional epidemiological changes to the data since then, particularly following the introduction of newer therapies for hyperkalaemia. Hence, it may be of interest to conduct this study using more recent data. Since our interest was strictly in studying the presence or absence of elevated potassium (hyperkalaemia), our reference group was comprised of both normokalaemic and hypokalaemic patients; however, it may be of interest in the future to study them separately, as others have shown differential mortality rates.13 Additionally, because the NIS is deidentified, it is possible that an individual may be present in the data more than once without means to identify such an occurrence; for that reason, the data are interpreted as independent hospital discharges, not as patients. Additionally, laboratory results are not available in the NIS. As such, the definition of hyperkalaemia in this study was based on its ICD code and limits our conclusions regarding potential causes of mortality, as the severity of hyperkalaemia is unknown. Hence, as with any study utilising ICD codes, our study may be subject to misclassification bias. Similarly, medications are not available in the NIS and we are unable to make inferences regarding the effects of therapies received before and/or after hyperkalaemia diagnosis. Finally, there were instances in which there was only one cluster within a stratum, so the SE could not be calculated; however, this happened in less than 1% of the data. While this work’s data source represents up to 97% of US hospital discharges, more work is needed to understand whether these findings generalise to other countries.

This post was originally published on