Académique Documents
Professionnel Documents
Culture Documents
DOI 10.1007/s00520-013-1734-6
ORIGINAL ARTICLE
Received: 25 August 2012 / Accepted: 28 January 2013 / Published online: 13 February 2013
# Springer-Verlag Berlin Heidelberg 2013
The Institutional Review Board at University of Texas M. The demographics and clinical characteristics were tabulated
D. Anderson Cancer Center (MDACC) approved the col- and compared with Chi-square test for categorical variables and
lection of the data and the current cross-sectional analysis. t test for continuous variables. The days of hospital stays were
The data were collected into an electronic database on all compared by ANOVAwith multiple comparisons adjusted with
patients admitted consecutively to MDACC for 3-months Tukey–Kramer method or Games–Howell test, the latter for
(May–July 2006). An admission was defined as a stay groups with unequal variance. Time to discharge was measured
of >23 h in the hospital that should also include midnight. from the date of hospital admission to the date of hospital
Data were checked at least once a week during this period discharge. Patients who died before discharge were considered
for accuracy and for inclusion of any missing data when censored at their dates of deaths. Time-to-90-day mortality was
available. The eligibility for inclusion in this analysis was measured from the date of admission to the date of death or last
any patient admitted to MDACC during this period with at follow-up at 90 days. The Kaplan–Meier product limit method
least one value for serum sodium at admission or during was used to estimate the survival outcomes of all patients by
hospital stay. For each patient, information on demograph- serum sodium groups, and comparisons among the groups
ics, medical conditions, laboratory data, treatments, clinical were achieved with the log-rank statistic. The natremia groups
outcomes, and billing data were electronically collected. were compared using Cox proportional hazard analyses for
Part of the data used here were previously analyzed, and a mortality. The proportional hazards assumption was tested with
paper on hyponatremia and another on acute kidney injury examination of Pearson correlation between Schoenfeld resid-
were recently published [6, 12]. The results of this hyper- uals and the rank of survival time for cases that progressed to an
natremia analysis have not been previously published. The event. After adjusting for other patient and clinical character-
primary outcome variable was hypernatremia defined as a istics, the model was fitted to determine the association of
serum sodium value of >147 mEq/L. The normal range of serum sodium levels with time-to-event outcomes. All results
serum sodium at MDACC laboratory is 135–147 mEq/L. from the Cox model are expressed in hazard ratios (HR) and
Among the 3,886 patients admitted, serum sodium values 95 % confidence intervals (CIs). For multivariate analyses,
were not available in 440 patients (11.3 %) yielding a total clinically relevant variables significant in the univariate analy-
of 3,436 patients for analysis. The data collected on their ses were included for adjustment in the final multivariate
first admission (i.e., n=3,436) were analyzed. To render the models. The relationship between serum sodium values and
large data of serum sodium values recorded during the survival was nonlinear. To overcome the nonlinearity, we
admission and hospital stay manageable for analysis, the employed restrictive cubic splines using four knots. The “post-
serum sodium values were extracted as values at “admis- rcspline package” for Stata version 10.0 was used to generate
sion,” “peak,” “low,” and “discharge.” For sensitivity anal- the curves and analyze the associations. All other survival
yses, multiple imputation method available in SPSS was analyses were carried out by using the SPSS (version 16.0;
used to impute the missing sodium values. All outcomes Chicago, IL). We also carried out multivariate regression anal-
were analyzed with and without imputed serum sodium ysis for hospital cost to compare the three groups for total cost
values, and the results were found to be not statistically of hospital stay. Age, type of malignancy, antibiotic and che-
different. The results presented here are based on data ex- motherapy administered, length of hospital stay, mean serum
cluding patients with missing serum sodium values. The creatinine, and mean hematocrit values were used as covariates.
accuracy and validity of the dataset used herein were con- Total cost of hospital stay was log transformed prior to analysis.
firmed by direct patients’ electronic chart checking that we
undertook to determine the possible causes for hypernatre-
mia and hospital death. Hypernatremia was defined as a Results
serum sodium value of >147 mEq/L at admission or any
time during hospital stay, and all patients with serum sodium From 1 May to 31 July of 2006, a total 3,886 patients were
values were included in the analysis. To assess the effect of admitted to MDACC, of which 75 % were admitted once,
hypernatremia on hospital cost and mortality, we obtained the 22 % for two to three times, and 3 % for more than three times.
final hospital bill for each patient obtained from the hospital For this analysis, we used data from the first admission on
database, and the data on patients’ mortality from the hospital 3,446 patients in whom serum sodium values were available.
cancer registry. Each patient with admission hypernatremia or
acquired hypernatremia during hospital stay (i.e., patients with Natremia breakdown and patient characteristics
normal serum sodium on admission but had peak serum
sodium above normal after admission) was also identified On admission, 69.9, 29.9, and 0.2 % were in the eunatremia,
and data between the two groups were compared. hyponatremia, and hypernatremia categories, respectively.
Support Care Cancer (2013) 21:1871–1878 1873
The respective values for the natremia breakdown for hospital the total hospital cost or bill was 45.7 % (95 % CI, 34.2–
stay (including admission) were 51.8, 45.6, and 2.6 % (Fig. 1) 57.9 %) higher in the hypernatremic than in eunatremic
indicating the development of new hyponatremia and hyper- patients (p<0.01) and was 36.9 % (95 % CI, 25.8–45.9 %)
natremia during patients’ stay in the hospital. On admission, higher than in hyponatremic patients (p<0.01) (Table 3).
only seven patients were hypernatremic which increased over The use of chemotherapy and antibiotic were also associated
10-fold to 90 patients during hospital stay. The clinical char- with significantly higher costs and so was as the longer
acteristics and demographics features of all patients and of hospital stay as would be expected (Table 3).
groups based on natremia breakdowns are given in Table 1. That hypernatremics had higher hospital bills even when
The prevalence of hematological malignancies, especially compared with hyponatremics is graphically depicted in
leukemia, was significantly higher in patients who had hyper- Fig. 4a.
natremia as compared with the rest (Table 1). The distribution
of the primary malignancies was: hematological (20.2 %), Mortality
genitourinary (13.5 %), gastrointestinal (14.5 %), head, neck,
and lung (16.2 %), and other (melanoma, breast, thyroid, The overall crude in-hospital mortality rate regardless of
unknown, or unidentified; 36.0 %). serum sodium was 5.4 %. When grouped by natremia
groups, this was 2.4, 7.3, and 39.3 % in the eunatremia,
Length of hospital stay hyponatremia, and hypernatremia groups (p<0.01 for all
comparisons). The causes of death in 35 hypernatremic
The overall length of stay of all patients regardless of serum patients who died in the hospital was multifactorial with
sodium values was 8.2±9.2 days (mean ± SD). The length infection as the most common denominator. Pneumonia
of stay of patients with hypernatremia was significantly followed by respiratory failure was noted in 37.0 % of the
higher at 26.7±22.3 days compared with 10.3±10.0 days patients; whereas, 17.4 % of the patients had sepsis due to
in patients with hyponatremia (p<0·01) and 5.6±4.9 days in infections other than pneumonia.
patients with eunatremia (p<0·01). That hypernatremics had The 90-day mortality and serum sodium values were plot-
longer hospital stay even when compared with hyponatre- ted as restrictive cubic splines, which showed a “U”-shaped
mics is graphically depicted in Fig. 4a. During the hospital relationship that was steeper for hypernatremia than for hypo-
stay, the percent of patients stayed in the critical care unit natremia (Fig. 2). The overall 90-day mortality rate was
was 10.9, 15.1, and 30.0 % for eunatremics, hyponatremics, 11.7 % and based on natremia groups were 7.5, 14.4, and
and hypernatremics, respectively (p<0.01 for all compari- 47.2 % in the eunatremia, hyponatremia, and hypernatremia
sons). That hypernatremics had the highest rate of ICU groups (p<0.01). Multivariate Cox regression analysis was
admissions even when compared with hyponatremics is carried out using the eunatremia group as reference category
graphically depicted in Fig. 4a. (Table 2). Adjusting for age, type of malignancy, chemother-
apy and antibiotic therapy, length of hospital stay, and serum
creatinine and hematocrit values, the HR for 90-day mortality
Hospital cost analysis in hypernatremics were significantly higher compared with
eunatremics (5.09 (95 % CI, 3.32–7.81); p<0·01). Repeating
In the multivariate regression model adjusted for age, type the analysis with hyponatremia as the reference category,
of malignancy, antibiotic and chemotherapy administered, analysis still showed a higher 90-day mortality in the hyper-
hospital stay, mean serum creatinine, and mean hematocrit, natremia group (HR, 2.79 (95 % CI, 1.91–4.11); p<0.01). The
Kaplan–Meier survival analysis also demonstrated signifi-
Hypernatremia cantly higher rates of mortality in hypernatremic than euna-
(>147 mEq/L)
2.6% tremic or hyponatremic patients (both p<0·01) (Fig. 3).
That hypernatremics had the highest rate of mortality
even when compared with hyponatremics is graphically
depicted in Fig. 4b.
Eunatremia
(135-147 mEq/L) Admission hypernatremia vs. hospital-acquired hypernatremia:
51.8% Hyponatremia patient characteristics and outcome analyses
(<135 mEq/L)
45.6%
Table 4 displays the comparison between the two groups.
Leukemic and stem cell transplant patients were more fre-
quent in the acquired groups, and most of them during admis-
Fig. 1 The natremia breakdowns in hospitalized cancer patients sion were receiving chemotherapy and were admitted to
1874 Support Care Cancer (2013) 21:1871–1878
Table 1 The patient characteristics: all patients, and groups based on natremia breakdowns
*p values (overall)
a
Include Asian, Filipino, Pacific Islander, and American Indian
b
Include leukemia, lymphoma, and myeloma
c
Include melanoma, breast, and thyroid malignancies
critical care unit. These suggest that the acquired hypernatre- MDACC over a 3-month period demonstrated that hyperna-
mic as a group were severely ill. Consistently, the length of tremia in hospitalized cancer patients was seen in 2.6 %, far
stay, hospital bills and the crude mortality rate were higher in less than the very high rates of 46 % of hyponatremia seen in
the acquired-hypernatremic group than the admission- concurrently admitted cancer patients. Most of the hyperna-
hypernatremic group, although mortality did not reach statis- tremia—nearly 90 %—were acquired in the hospital and
tical significance (Table 4). The sample size of patients in the hypernatremia although was infrequent, however, associated
admission-hypernatremia group was quite small and may not with unexpectedly worse clinical outcomes than hyponatre-
fully represent the true population. mia, i.e., higher mortality, hospital stay, and hospital bills.
In a prospective study of patients admitted to general
medical wards of a large urban university hospital for
Discussion 3 months, hypernatremia as defined as >150 mEq/L of serum
sodium was noted in 1 % of patients [8]. Although no com-
Our cross-sectional analysis of data collected from a large parison to eunatremic or hyponatremic patients were provid-
number of patients admitted consecutively to the UT ed, the mortality rate of 41 % reported for patients with
Fig. 2 The restrictive cubic
Predicted probability of 90-day mortality
.8
.6
.4
.2
.2
0
0
110 120 130 140 150 160 110 120 130 140 150 160
Mean Serum sodium (mEq/L) Mean Serum sodium (mEq/L)
Support Care Cancer (2013) 21:1871–1878 1875
Table 2 Multivariate hazard ratios (HR) for in-hospital and 90-day mortality showing significant association with hypernatremia
Multivariate
Age (year) 0.99 ( 0.98–1.01) 0.49 1.01 (1.001–1.02) <0.01
Malignancies (hematological vs. non-hematological) 1.16 (0.78–1.74) 0.47 1.61 (1.22–2.12) <0.01
Chemotherapy (yes vs. no) 0.49 (0.34–0.72) <0.01 1.10 (0.87–1.42) 0.42
Antibiotic use (yes vs. no) 0.61 ( 0.39–0.94) 0.02 1.17 (0.90–1.52) 0.24
Hospital stay in days – – 1.002 (0.99–1.01) 0.36
Serum creatininea 1.25 (1.12–1.40) <0.01 1.29 (1.18–1.41) <0.01
Hematocrita 0.91 (0.87–0.95) <0.01 0.96 (0.93–0.98) 0.01
Serum sodium (categories)
Eunatremia (135–147 mEq/L) 1.00 (ref) 1.00 (ref)
Hyponatremia (<135 mEq/L) 1.21 ( 0.83–1.74) 0.34 1.81 (1.44-2.29) <0.01
Hypernatremia (>147 mEq/L) 2.17 ( 1.30–3.64) <0.01 5.09 (3.32–7.81) <0.01
a
Mean of all values recorded during hospital stay
hypernatremic in this study suggest that mortality in this Although the data on outcomes provided were limited, mor-
subgroup is likely much higher than non-hypernatremic tality rates among the hypernatremic patients were over 2-fold
patients. In another study of 8,142 consecutive adults admitted higher than the hyponatremic patients [9]. Thus in four set-
to a medical–surgical intensive care unit (ICU) between 2000 tings, ICU, general medical hospital, ER, and our comprehen-
and 2006, the frequency of ICU-acquired hypernatremia in- sive cancer center, hypernatremia was found to be consistently
terestingly was higher than hyponatremia (26 vs. 11 %), but associated with higher mortality even when compared against
important to this discussion, hypernatremia was associated patients with hyponatremia.
with higher hospital mortality compared with hyponatremia Hyponatremia is known to be associated with higher cost,
(34 vs. 28 %) [7]. A recent analysis based on retrospective but there are no published studies to our knowledge on the
case series of 43,911 patients with serum sodium seen at a impact of hypernatremia on hospital cost. Our study dem-
university hospital’s Emergency Room (ER) in Switzerland, onstrated that even after adjusting for several factors likely
hypernatremia was noted in 2 % and hyponatremia in 10 %. to affect the cost of patients care in the hospital (Table 3),
Days
Number at risk
Eunatremia 1785 1756 1733 1719 1712 1703 1647
Hyponatremia 1571 1489 1445 1426 1408 1396 1340
Hypernatremia 90 81 69 65 61 58 47
1876 Support Care Cancer (2013) 21:1871–1878
500 Cost of stay stem cell transplant service. We speculate that hospital-
Three outcomes expressed as percent Critical care stay acquired hypernatremia may likely be related to the use of
Length of stay
diuretics and chemotherapy and the practice of keeping
400 patients “dry” to avoid pulmonary edema, such as in patients
of same fo eunatremia (%)
Multivariate regression
Age (years) 0.001 (−0.001–0.002) 0.24
Malignancies (hematological vs. non-hematological) −0.060 (−0.116–-0.005) 0.03
Chemotherapy (yes vs. no) 0.242 (0.194–0.289) <0.01
Antibiotic use (yes vs. no) 0.327 (0.283–0.370) <0.01
Hospital stay in days 0.055 (0.052–0.057) <0.01
Serum creatininea 0.005 (−0.026–0.035) 0.76
Hematocrita −0.002 (−0.006–0.002) 0.38
a
Mean of all values during hos- Serum sodium (categories)
pital stay. In this multivariate
model, hypernatremia had strong Eunatremia (135–147 mEq/L) 0.00 (ref)
correlation to cost; the cost in Hyponatremia (<135 mEq/L) 0.094 (0.054–0.134) <0.01
hypernatremics was nearly 46 % Hypernatremia (>147 mEq/L) 0.457 (0.342–0.571) <0.01
higher than eunatremics
sensitivity analyses, employing restrictive cubic splines for covariates—known or unknown—as well as comorbidities
nonlinear data, and multivariate regression analysis for cost and their severity into the regression models could have led
analysis. A main limitation of our study is lack of details on to some unmeasured confounding in the multivariate analy-
the causes of hypernatremia in this population as alluded to. ses. However, we had included common variable relevant to
Prospective studies are required in cancer patients to deter- the survival of cancer population.
mine the nature and cause of evolution of hypernatremia and In summary, we provide what we believe is the first
to determine whether treatment to mitigate hypernatremia report on a detailed outcome analysis of hypernatremia in
will change the outcomes. Our inability to incorporate all hospitalized cancer patients. Hypernatremia was seen in
2.6 % of patients compared with 46 % of patients with 2. Schrier RW (2008) Vasopressin and aquaporin 2 in clinical disor-
ders of water homeostasis. Semin Nephrol 28(3):289–296
hyponatremia. Most hypernatremia were acquired in the
3. Anderson RJ (1986) Hospital-associated hyponatremia. Kidney Int
hospital. Hypernatremia was seen much more commonly 29(6):1237–1247
in critically ill patients especially in patients with leukemia 4. Upadhyay A, Jaber BL, Madias NE (2009) Epidemiology of
and stem cell transplants. Patients with hypernatremia had hyponatremia. Semin Nephrol 29(3):227–238
5. Waikar SS, Mount DB, Curhan GC (2009) Mortality after hospi-
substantially worse clinical outcomes and higher health care
talization with mild, moderate, and severe hyponatremia. Am J
cost than eunatremic or even hyponatremic patients (Fig. 4). Med 122(9):857–865
If our findings are to be confirmed in prospective studies in 6. Doshi SM, Shah P, Lei X, Lahoti A, Salahudeen AK (2012)
cancer or noncancer patients, future studies are warranted to Hyponatremia in hospitalized cancer patients and its impact on
clinical outcomes. Am J Kidney Dis: Offic J National Kidney
examine whether avoidance or prompt and sustained cor- Found 59(2):222–228
rection of hypernatremia could lead to a better outcome. 7. Stelfox HT, Ahmed SB, Khandwala F, Zygun D, Shahpori R,
Laupland K (2008) The epidemiology of intensive care unit-
acquired hyponatraemia and hypernatraemia in medical–surgical
intensive care units. Crit Care 12(6):R162
Conflict of interest statement We are submitting herewith the man- 8. Palevsky PM, Bhagrath R, Greenberg A (1996) Hypernatremia in
uscript titled “The frequency, cost, and clinical outcomes of hypernatre- hospitalized patients. Ann Intern Med 124(2):197–203
mia in patients hospitalized to a comprehensive cancer center”—to be 9. Arampatzis S, Frauchiger B, Fiedler GM, Leichtle AB, Buhl D,
considered for publication as an original article in Supportive Care in Schwarz C et al (2012) Characteristics, symptoms, and outcome of
Cancer. There are no financial relationships to disclose for any of the severe dysnatremias present on hospital admission. The American
authors. The manuscript has not been published elsewhere nor is it being journal of medicine 125(11):1125, e1–e7
considered for publication. We have full control of all primary data, and 10. Salahudeen AK, Bonventre JV (2013) Onconephrology: the latest
the journal is free to review these data if required. frontier in the war against kidney disease. J Am Soc Nephrol:
JASN 24(1):26–30
11. Elsayem A, Mori M, Parsons HA, Munsell MF, Hui D, Delgado-
Guay MO et al (2010) Predictors of inpatient mortality in an acute
palliative care unit at a comprehensive cancer center. Support Care
References Cancer 18(1):67–76
12. Salahudeen AK, Doshi SM, Pawar T, Nowshad G, Lahoti A, Shah
P (2013) Incidence rate, clinical correlates, and outcomes of AKI
1. Adrogue HJ, Madias NE (2000) Hypernatremia. N Engl J Med 342 in patients admitted to a comprehensive cancer center. Clin J Am
(20):1493–1499 Soc Nephrol. doi:10.2215/CJN.03530412