Restrictive cubic spline depicting the unadjusted relationship between hospital admission serum sodium concentrations and in-hospital mortality. Dashed lines represent the 95% confidence interval. To convert serum sodium concentration to millimoles per liter, multiply by 1.0.
Adjusted relationships between community-acquired hyponatremia (admission serum sodium concentration [Na+], <138 mEq/L [to convert to millimoles per liter, multiply by 1.0.]) (A) and hospital-acquired hyponatremia (admission serum [Na+] level, 138-142 mEq/L, and nadir [Na+], <138 mEq/L) (B) and in-hospital mortality in selected demographic and clinical subgroups. Some subgroups represented in part A are not described in part B owing to the low number of deaths in these subgroups. CI indicates confidence interval; eGFR, estimated glomerular filtration rate.
Wald R, Jaber BL, Price LL, Upadhyay A, Madias NE. Impact of Hospital-Associated Hyponatremia on Selected Outcomes. Arch Intern Med. 2010;170(3):294-302. doi:10.1001/archinternmed.2009.513
Hyponatremia is the most common electrolyte disorder encountered in hospitalized patients.
We evaluated whether hospital-associated hyponatremia has an independent effect on all-cause mortality, hospital length of stay (LOS), and patient disposition. This cohort study included all adult hospitalizations at an academic medical center occurring between 2000-2007 for which an admission serum sodium concentration ([Na+]) was available (N = 53 236). We examined community-acquired hyponatremia (admission serum [Na+], <138 mEq/L [to convert to millimoles per liter, multiply by 1.0]), hospital-aggravated hyponatremia (community-acquired hyponatremia complicated by worsening in serum [Na+]), and hospital-acquired hyponatremia (nadir serum [Na+], <138 mEq/L with a normal admission serum [Na+]). The independent associations of these hyponatremic presentations with in-hospital mortality, LOS, and patient disposition were evaluated using generalized estimating equations adjusted for age, sex, race, admission service, and Deyo-Charlson Comorbidity Index score.
Community-acquired hyponatremia occurred in 37.9% of hospitalizations and was associated with adjusted odds ratios (ORs) of 1.52 (95% confidence interval [CI], 1.36-1.69) for in-hospital mortality and 1.12 (95% CI, 1.08-1.17) for discharge to a short- or long-term care facility and a 14% (95% CI, 11%-16%) adjusted increase in LOS. Hospital-acquired hyponatremia developed in 38.2% of hospitalizations longer than 1 day in which initial serum [Na+] was 138 to 142 mEq/L. Hospital-acquired hyponatremia was associated with adjusted ORs of 1.66 (95% CI, 1.39-1.98) for in-hospital mortality and 1.64 (95% CI, 1.55-1.74) for discharge to a facility and a 64% (95% CI, 60%-68%) adjusted increase in LOS. The strength of these associations tended to increase with hyponatremia severity.
Hospital-associated hyponatremia is a common occurrence. All forms of hyponatremia are independently associated with in-hospital mortality and heightened resource consumption.
Hyponatremia is the most common electrolyte disorder encountered in hospitalized patients. Many studies1- 8 have addressed the epidemiologic aspects of hospital-associated hyponatremia, including its adverse prognostic effect on patient outcomes. However, these studies have variably focused on either admission hyponatremia (community-acquired hyponatremia [CAH])4- 6 or hyponatremia that developed after admission (hospital-acquired hyponatremia [HAH]).3,7 Other studies restricted the evaluation of hospital-associated hyponatremia to more severe degrees of the disorder2,8- 10 or to specific medical conditions.11- 17 Only a few studies have evaluated the impact of hyponatremia on clinical outcomes other than mortality, such as hospital length of stay (LOS),9,13,16,17 and, to our knowledge, none has explored the potential impact of hyponatremia on patient disposition.
In the present study, we characterized the entire spectrum of hospital-associated hyponatremia in a large unselected group of adults who received care at an urban academic medical center (St Elizabeth's Medical Center, Boston, Massachusetts). These analyses encompassed CAH and HAH. Among patients with CAH, we also distinguished a large cohort of patients in whom the hyponatremia worsened during hospitalization (hospital-aggravated hyponatremia). We evaluated the associations between each of these hyponatremic presentations and important hospitalization outcomes, including in-hospital mortality, LOS, and patient disposition, while adjusting for key confounders of these relationships.
All discharges of adult patients from St Elizabeth's Medical Center, a 400-bed acute care tertiary hospital, between October 1, 2000, and September 30, 2007, were considered for inclusion in this analysis. Availability of a serum sodium concentration ([Na+]) at the time of admission (defined as the actual day of admission or 1 day earlier) was necessary to classify each hospitalization by the appropriate hyponatremic presentation. In all of the assessments, the hospitalization, rather than the patient, was the unit of analysis.
Discharge abstracts provided information on patient age, sex, race, admission service, and up to 15 International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnostic codes. The discharge abstract file was linked with the hospital's electronic laboratory database, from which we extracted serum [Na+] values for the corresponding hospitalization. Owing to the confounding effect on serum [Na+] of water translocated from cells to the extracellular fluid in hyperglycemia, all serum [Na+] values in this analysis were adjusted for concomitantly measured serum glucose levels. If the glucose level was greater than 100 mg/dL (to convert to millimoles per liter, multiply by 0.0555), serum [Na+] values were adjusted upward by 2 mEq/L (to convert to millimoles per liter, multiply by 1.0) for each 100-mg/dL increment in serum glucose.18,19
In a restrictive cubic spline20 constructed to ascertain the crude relationship between admission serum [Na+] and mortality, a U-shaped relationship was observed, with a value of 140 mEq/L associated with the lowest risk of mortality (Figure 1). On the basis of this finding, we evaluated the unadjusted relationships between 1-mEq/L changes in serum [Na+], using 140 mEq/L as the reference value. We found that mortality was significantly associated with serum [Na+] less than 138 mEq/L and greater than 142 mEq/L. Therefore, 138 to 142 mEq/L was designated as the reference range for normonatremia in this study.
Community-acquired hyponatremia was defined as a serum [Na+] value less than 138 mEq/L at the time of hospital admission. Hospital-aggravated hyponatremia was designated when a further decline in the serum [Na+] value of at least 2 mEq/L occurred during the first 48 hours of a CAH-associated hospitalization. Hospital-acquired hyponatremia was defined as development of a nadir serum [Na+] level less than 138 mEq/L during hospitalization in the setting of a serum [Na+] value within the reference range at the time of admission. We assessed the impact of graded degrees of severity of CAH and HAH by classifying serum [Na+] values in decrements of 5 mEq/L below 138 mEq/L. Finally, for CAH and HAH, we conducted secondary analyses in which the serum [Na+] value was evaluated as a continuous variable below 138 mEq/L.
In-hospital mortality, LOS, and patient disposition were each evaluated. For disposition status, the event of interest was discharge to a short- or long-term care facility; discharges to home and departures from the hospital against medical advice comprised the reference category. Hospitalizations during which the patient died were excluded from analyses examining LOS and disposition status.
The following covariates were considered: age, sex, race (white, black, Asian, other, or missing), admission service (medical, surgical, or other), and the Deyo-Charlson Comorbidity Index.21,22 In mortality and disposition status analyses, age was treated as a piecewise linear variable with the cutoff point at age 60 years to account for the much steeper slope seen for individuals older than 60 years. Piecewise linear regression allows for 2 different slopes to be observed, one for patients 60 years or younger and the other for patients older than 60 years. Age was treated as a continuous variable in the LOS analysis. The Deyo-Charlson Comorbidity Index incorporates a patient's history of comorbidities using ICD-9-CM diagnosis codes, with increasing numerical values reflecting greater comorbidity. For each hospitalization, the Deyo-Charlson Comorbidity Index was used to generate a score, and hospitalizations were categories based on scores of 0 through 3 or higher.
To explore the consistency of the findings in various segments of the population, we evaluated the adjusted impact of CAH and HAH on in-hospital mortality in various subgroups defined by demographic profiles, admission services, and comorbidities. Individual comorbidities were defined using the ICD-9-CM diagnosis codes used in the Deyo-Charlson Comorbidity Index. We also evaluated the impact of hyponatremia within strata of kidney function. The latter were categorized according to prevailing recommendations23 using estimated glomerular filtration rates calculated on the basis of the 4-variable Modification of Diet in Renal Disease equation.24 To avoid model overfitting, we reported adjusted effect estimates only if at least 50 deaths occurred in the subgroup of interest.
Continuous variables are described as means (SDs) or medians (25th to 75th percentiles), as appropriate. Categorical variables are expressed as percentages. The final analytic data set included 53 236 hospital admissions involving 29 904 patients. We presented the characteristics of hospitalizations with and without associated CAH and HAH. To calculate the difference and 95% confidence interval (CI), we used linear regression with generalized estimating equations (GEEs) to adjust for clustering of multiple admissions within patients. Given the large sample size, we invoked the central limit theorem to assume normality for binary outcomes. The independent effects of hyponatremia on in-hospital mortality, LOS, and discharge to a short- or long-term care facility were evaluated after adjusting for age, sex, race, admission service, and comorbidities using GEEs. Given the clustering of multiple admissions among certain patients, GEEs permit the inclusion of all admissions into the analysis while generating the correct standard errors for use in hypothesis testing. In-hospital mortality and discharge to a facility were modeled using the GENMOD procedure in SAS (SAS Institute Inc, Cary, North Carolina) using a binomial distribution, the logit link, and the autoregressive covariance structure. The LOS was evaluated using repeated-measures GEEs with the negative binomial distribution, log link, and autoregressive covariance structure. Negative binomial models are useful for count data when the variability of the outcome is greater than that observed in a Poisson distribution. The effect estimate generated by this analysis reflects the proportional change in the LOS compared with the reference category.
Finally, to further limit the influence of patients with recurrent admissions on the calculated effect estimates, the relationships between the various hyponatremic presentations and in-hospital mortality were modeled using logistic regression that included only the first hospitalization per patient.
The functional form of serum [Na+] and its association with in-hospital mortality was examined using restricted cubic splines with 4 knots. Plots of the restricted splines were constructed using the design library of R version 2.6 (Free Software Foundation, Boston). All other analyses and plots were generated using SAS version 9.1 (SAS Institute Inc).
There were 97 472 hospitalizations during the 7-year study. Obstetrical admissions (n = 11 589) were not analyzed owing to the expected presence of mild physiologic hyponatremia in pregnancy.25 Other reasons for exclusion included absence of [Na+] value on the day of admission (n = 16 035), absence of any [Na+] values during hospitalization (n = 6030), inability to determine admission serum [Na+] level (n = 927), missing service assignment (n = 69), ambiguous discharge plan (n = 2), and admission serum [Na+] level greater than 142 mEq/L (n = 9584). The final analytic data set thus comprised 53 236 hospital admissions.
Community-acquired hyponatremia, as defined by an admission serum [Na+] less than 138 mEq/L, was identified in 20 181 hospitalizations (37.9%). Patients with CAH were older, had a higher comorbidity score, and were more likely to be admitted to medical services (Table 1). Community-acquired hyponatremia was associated with higher in-hospital mortality (3.4% vs 2.0%) and a higher likelihood of discharge to a short- or long-term care facility (37.5% vs 33.7%).
In addition, CAH was associated with an adjusted odds ratio (OR) of 1.52 (95% CI, 1.36-1.69) for in-hospital mortality (Table 2). When analyzed by strata of admission values, even mild hyponatremia (serum [Na+], 133-137 mEq/L) was independently associated with mortality (adjusted OR, 1.34; 95% CI, 1.18-1.51), and the risk of death tended to increase as hyponatremia worsened. When evaluated as a continuous variable, each 1-mEq/L decline in serum [Na+] value below 138 mEq/L was associated with an 8% increase in the risk of death (OR, 1.08; 95% CI, 1.07-1.10). Also, CAH was associated with a heightened likelihood of discharge to a short- or long-term care facility (OR, 1.12; 95% CI, 1.08-1.17) and a 14% (95% CI, 11%-16%) relative prolongation in LOS (Table 2).
The impact of CAH was consistent across a wide range of subgroups (Figure 2A). The adjusted risk of mortality was more prominent in hospitalizations involving patients younger than 65 years (OR, 2.58; 95% CI, 1.98-3.35) vs patients 65 years and older (1.44; 1.28-1.63) (P < .001 for interaction). The association between CAH and mortality tended to be stronger on admissions to surgical services (adjusted OR, 2.21; 95% CI, 1.49-3.29) compared with those involving medical services (1.52; 1.36-1.71), but the interaction did not reach significance (P = .08).
Aggravation of hyponatremia occurred in 1151 hospitalizations (5.7%) where hyponatremia was present at the time of admission. Compared with hospitalizations with no CAH, hospital-aggravated hyponatremia was independently associated with a higher risk of in-hospital mortality (OR, 2.30; 95% CI, 1.75-3.02), whereas the presence of CAH with no further decline in serum [Na+] level was associated with an adjusted OR of 1.46 (95% CI, 1.31-1.64). In addition, hospital-aggravated hyponatremia was associated with a prolonged LOS and a higher likelihood of discharge to a facility (Table 3).
To evaluate HAH, we restricted the cohort to hospitalizations in which the initial serum [Na+] level was within the reference range (138-142 mEq/L) and LOS was longer than 1 day (n = 27 897). Hospital-acquired hyponatremia developed in 10 662 of these hospitalizations (38.2%). Patients with HAH were older and were more likely to be admitted to surgical services and to have a higher comorbidity score (Table 4). Almost all the comorbidities were more prevalent in hospitalizations with HAH. Hospital-acquired hyponatremia was associated with higher in-hospital mortality (2.9% vs 1.4%). Among survivors of hospitalization, an increase in LOS (median, 5 vs 4 days) and a higher likelihood of discharge to a short- or long-term facility (45.7% vs 31.9%) were observed.
Hospital-acquired hyponatremia was associated with a 66% increase in the adjusted odds of in-hospital mortality (OR, 1.66; 95% CI, 1.39-1.98) (Table 5). This relationship was observed for hospitalizations with mild HAH (nadir [Na+], 133-137 mEq/L; OR, 1.31; 95% CI, 1.08-1.58), and it progressively strengthened with worsening hyponatremia (Table 5). When serum [Na+] was evaluated as a continuous variable, the adjusted risk of death was increased by 23% (OR, 1.23; 95%, 1.19-1.27) for each 1-mEq/L decline below 138 mEq/L. Hospital-acquired hyponatremia was associated with an adjusted OR of 1.64 (95% CI, 1.55-1.74) for discharge to a short- or long-term care facility and a 64% (95% CI, 60%-68%) adjusted increase in LOS (Table 5). The strength of these associations also increased with the severity of hyponatremia (Table 5). The association between HAH and in-hospital mortality was consistent across most subgroups (Figure 2B).
To mitigate the effect of patients with recurrent hospitalizations on the results, we performed sensitivity analyses that considered only the first hospitalization for each individual. Patients with CAH (adjusted OR, 1.56; 95% CI, 1.34-1.83) and HAH (1.61; 1.29-2.01) had a higher risk of in-hospital mortality. These effect estimates were similar to those obtained in the primary analyses that included all hospitalizations.
This analysis of more than 50 000 admissions to St Elizabeth's Medical Center revealed that hospital-associated hyponatremia is a common occurrence with important consequences. Whether it is present on admission, exacerbated after admission, or develops during hospitalization, hyponatremia is independently associated with in-hospital mortality, prolongation of LOS, and discharge to a facility.
In this unselected population of hospitalized adults, CAH and HAH each developed in approximately 38% of at-risk hospitalizations. These high rates of hyponatremia substantially exceed those reported by other researchers.4,5,26 This is largely attributable to the more liberal definition of hyponatremia that we used. Although 135 to 145 mEq/L is frequently used as the reference range for serum [Na+], the present data revealed an increase in mortality as serum [Na+] values declined below 138 mEq/L or rose above 142 mEq/L, justifying the adoption of an operational definition of normonatremia of 138 to 142 mEq/L. Even serum [Na+] values that would conventionally be classified as normal or slightly below normal (ie, 133-137 mEq/L) were independently associated with mortality, prolonged LOS, and discharge to a facility. The association between mortality and serum [Na+] values in ranges that would conventionally be perceived as low normal has also been observed in the setting of heart failure13,17 and liver disease27 but never before in an unselected population of hospitalized adults. The present data suggest that the conventionally accepted reference range of serum [Na+] level in hospitalized patients may need reconsideration. Further examination of this question in multiple centers and in a variety of clinical settings would be required before the current reference range is altered.
Hyponatremia in unselected adults who survived hospitalization was independently associated with prolongation of LOS and discharge to a short- or long-term care facility. In CAH and HAH, these associations became stronger with the severity of hyponatremia. The implication is that hyponatremia imparts a considerable burden on the consumption of health care resources, as has been suggested by other researchers.28,29
Postadmission aggravation of hyponatremia, using a relatively restrictive definition, was observed in 6% of CAH-associated hospitalizations, and HAH developed in 38% of normonatremic admissions. Most hyponatremia is dilutional in nature, with water retention most commonly resulting from a water-excreting defect coupled with water intake exceeding the renal excretory capacity plus insensible losses.30 Such water-excreting defects can precede admission or can worsen or develop during hospitalization owing to antidiuretic effects, including medications, pain, nausea, organ failure, and the postoperative state.8,30,31 A water-excreting defect notwithstanding, hyponatremia will not worsen or develop unless positive water balance ensues. The present data emphasize the need for vigilance when prescribing fluids, particularly hypotonic fluids, and medications with antidiuretic activity to hospitalized patients.
Severity of hyponatremia prognosticates adverse outcomes, especially when hyponatremia develops in the hospital. Although the adverse outcomes of severe hyponatremia (defined as [Na+] value <125 mEq/L) have been reported by other researchers,8- 10 some studies9,10 did not differentiate between hyponatremia at presentation and that developing in the hospital. In patients with HAH, a 15-fold increase in the risk of death was observed with nadir serum [Na+] of 127 mEq/L or less. In comparison, ORs for mortality were more modest in patients with the most severe forms of CAH. Differences in the pace of development of hyponatremia between CAH and HAH might contribute to the difference in risk. Moreover, compared with CAH, HAH was shown to be frequently unrecognized, hence being associated with delays in the initiation of treatment.8 Alternatively, patients with HAH might have developed severe illness or complications thereof in the hospital, whereas the causes associated with CAH might have been comparatively benign. These and other factors might explain the more ominous outcomes associated with HAH in this cohort.
The robust nature of these global findings is supported by the fact that the risk of death associated with hyponatremia was evident in almost every subgroup analyzed. The observation that the impact of CAH and HAH on mortality was accentuated in hospitalizations involving patients younger than 65 years is intriguing and requires further investigation.
This study has several strengths. We studied a large, diverse, and unselected population with an array of comorbid conditions. In addition to discerning between CAH and HAH in a cohort encompassing the entire hyponatremic spectrum, this is the first study, to our knowledge, to provide insights into the hospital aggravation of CAH. We also assessed the impact of hyponatremia on surviving patients by examining LOS and discharge disposition. Finally, the use of GEEs permitted analysis of all evaluable hospitalizations while reliably accounting for clustering of patients with multiple admissions.
There are also several important limitations to consider. The identification of individual comorbidities using administrative codes is relatively crude and does not account for the severity of the condition in question. The inability to incorporate the severity of these comorbidities might have led to some unmeasured confounding in the multivariable analyses. However, in the principal models relating hyponatremia and adverse outcomes, case mix was considered by the use of a well-validated instrument21,22 that summarized the global burden of illness associated with each hospitalization. Ascertainment of the end points also has some drawbacks. We were restricted to evaluating in-hospital mortality and, hence, were unable to capture events that occurred after discharge. Although access to postdischarge survival data would have optimized evaluation of the relationship between hyponatremia and mortality, this relationship has previously been shown to persist after discharge.4,15 We also were unable to differentiate disposition status between patients who presented to the hospital from short- or long-term care facilities and those who presented from home. Thus, it is unclear whether discharges to facilities reflected a deterioration in clinical status or whether some patients were merely returning to their preadmission residence. Because this study was conducted at a single hospital, the generalizability of these findings may be impaired. In particular, disposition plans and LOS might be susceptible to practices and prevailing conditions in this hospital and local health care system and might not be extrapolated to other settings. Furthermore, these findings were generated using data from hospitalized patients and cannot be extended to the ambulatory setting. Finally, although the spline curve suggested a progressive increase in mortality risk as admission serum [Na+] level declined, the few patients with an admission serum [Na+] value of less than 123 mEq/L should lead to a cautious interpretation of effect estimates for hospitalizations with serum [Na+] below this level.
Hospital-associated hyponatremia is a common occurrence, and all hyponatremic presentations and any level of hyponatremia portend adverse outcomes in hospitalized patients. The neurologic morbidity and mortality that arise from severe hypotonic hyponatremia and its overly rapid correction are well recognized.30,32,33 Even mild, seemingly asymptomatic, hyponatremia was recently found to be associated with attention impairment, gait instability, and falls.34 However, the key question is whether hyponatremia in most patients is simply a powerful marker of the severity of the underlying condition(s) that led to its development or a direct contributor to the adverse outcomes observed. It is conceivable that hyponatremia itself, the associated hypotonicity, or the elevated levels of arginine vasopressin might exert adverse effects on the cardiovascular and other organ systems, with important implications for patient morbidity and mortality. As a corollary, it remains unclear whether biochemical rectification of hyponatremia will confer improved patient outcomes. The advent of arginine vasopressin receptor antagonists presents an opportunity to test this important clinical question in patients with euvolemic or hypervolemic hyponatremia.35 Whether the relationship between hyponatremia and adverse outcomes is causal or associative, hyponatremia is a compelling prognostic marker of adverse outcomes. The identification of even mild hyponatremia should compel physicians to exercise heightened vigilance. Recognition of the high risk associated with all forms of this electrolyte abnormality should stimulate measures to prevent its development, but once present, close patient monitoring and careful management are mandatory.
Correspondence: Nicolaos E. Madias, MD, Department of Medicine, St Elizabeth's Medical Center, 736 Cambridge St, Boston, MA 02135 (email@example.com).
Accepted for Publication: October 16, 2009.
Author Contributions:Study concept and design: Wald, Jaber, and Madias. Acquisition of data: Jaber and Madias. Analysis and interpretation of data: Wald, Jaber, Price, Upadhyay, and Madias. Drafting of the manuscript: Wald and Madias. Critical revision of the manuscript for important intellectual content: Jaber, Price, Upadhyay, and Madias. Statistical analysis: Wald and Price. Administrative, technical, and material support: Jaber and Madias. Study supervision: Jaber and Madias.
Financial Disclosure: Dr Madias has served as a consultant to Astellas and Sanofi-Aventis and has received lecture fees from Astellas. Dr Jaber has received grant support from Genzyme Corp.
Funding/Support: Dr Wald is supported by the Randomized Controlled Trial Mentoring Program of the Canadian Institutes of Health Research and by an unrestricted fellowship award from Amgen Inc. Dr Jaber is supported by grants DK065102 and R03DK07775 from the National Institutes of Health.
Previous Presentation: These findings were presented in part as a scientific poster at the American Society of Nephrology Renal Week; November 16, 2006; San Diego, California.