Clinical Nutrition 36 (2017) 1391e1396

Contents lists available at ScienceDirect

Clinical Nutrition
journal homepage: http://www.elsevier.com/locate/clnu

Original Article

Costs of hospital malnutrition

Lori Jane Curtis a, *, Paule Bernier b, Khursheed Jeejeebhoy c, Johane Allard d,
Donald Duerksen e, Leah Gramlich f, Manon Laporte g, Heather H. Keller h
a
Department of Economics, University of Waterloo, 200 University Ave, Waterloo, ON, N2L 3G1, Canada
b
Jewish General Hospital, 3755 ch Cote Ste-Catherine, Montreal, QC, H3T 1E2, Canada
c
Department of Medicine, St-Michael Hospital, University of Toronto, 30 Bond Street, Toronto, ON, M5B 1W8, Canada
d
Department of Medicine, University Health Network, University of Toronto, 585 University Avenue, 9N-973, Toronto, ON, M5G 2C4, Canada
e
Department of Medicine, St-Boniface Hospital, University of Manitoba, 409 Tache Avenue, Winnipeg, MB, R2H 2A6, Canada
f
Department of Medicine, University of Alberta, Alberta Health Services, Community Services Centre, Royal Alexandra Hospital, Edmonton, AB, T5H 3V9,
Canada
g 
Reseau de Sant
e Vitalit
e Health Network, Campbelton Regional Hospitals, 189 Lily Lake Road, PO Box 880, Campbellton, NB, E3N 3H3, Canada
h
Schlegel-UW Research Institute for Aging, University of Waterloo, 200 University Avenue West, Waterloo, ON, N2L 3G1, Canada

a r t i c l e i n f o s u m m a r y

Article history: Background & aims: Hospital malnutrition has been established as a critical, prevalent, and costly
Received 18 February 2016 problem in many countries. Many cost studies are limited due to study population or cost data used. The
Accepted 13 September 2016 aims of this study were to determine: the relationship between malnutrition and hospital costs; the
influence of confounders on, and the drivers (medical or surgical patients or degree of malnutrition) of
Keywords: the relationship; and whether hospital reported cost data provide similar information to administrative
Malnutrition
data. To our knowledge, the last two goals have not been studied elsewhere.
Hospital costs
Methods: Univariate and multivariate analyses were performed on data from the Canadian Malnutrition
Hospital outcomes
Subjective global assessment
Task Force prospective cohort study combined with administrative data from the Canadian Institute for
Health Information. Subjective Global Assessment was used to assess the relationship between nutri-
tional status and length of stay and hospital costs, controlling for health and demographic characteristics,
for 956 patients admitted to medical and surgical wards in 18 hospitals across Canada.
Results: After controlling for patient and hospital characteristics, moderately malnourished patients'
(34% of surveyed patients) hospital stays were 18% (p ¼ 0.014) longer on average than well-nourished
patients. Medical stays increased by 23% (p ¼ 0.014), and surgical stays by 32% (p ¼ 0.015). Costs
were, on average, between 31% and 34% (p-values < 0.05) higher than for well-nourished patients with
similar characteristics. Severely malnourished patients (11% of surveyed patients) stayed 34% (p ¼ 0.000)
longer and had 38% (p ¼ 0.003) higher total costs than well-nourished patients. They stayed 53%
(p ¼ 0.001) longer in medical beds and had 55% (p ¼ 0.003) higher medical costs, on average. Trends
were similar no matter the type of costing data used.
Conclusions: Over 40% of patients were found to be malnourished (1/3 moderately and 1/10 severely).
Malnourished patients had longer hospital stays and as a result cost more than well-nourished patients.
© 2016 Elsevier Ltd and European Society for Clinical Nutrition and Metabolism. All rights reserved.

1. Introduction increased costs associated with malnutrition during hospitaliza-

tion. However, our understanding of the costs attributed to
Hospital malnutrition [1] (inadequate intake of energy, protein malnutrition are limited due to the use of single clinical settings
and nutrients) has been established as a critical, prevalent, and [5,6,8], single clinical populations, such as those with cancer [7],
costly problem for patients and the health care system. Studies in retrospective studies [7,9], or hypothetical samples based on
Europe [2e7], Asia [8], and South America [9] have demonstrated costing and nutritional status estimates from population level data
[4]. Few studies to date have been multi-center, limiting the
generalizability of prevalence estimates of malnutrition and the
* Corresponding author. costs associated with impaired nutritional states [2,9].
E-mail address: ljcurtis@uwaterloo.ca (L.J. Curtis).

http://dx.doi.org/10.1016/j.clnu.2016.09.009
0261-5614/© 2016 Elsevier Ltd and European Society for Clinical Nutrition and Metabolism. All rights reserved.
1392 L.J. Curtis et al. / Clinical Nutrition 36 (2017) 1391e1396

Additionally, nutritional status designation in these analyses is Demographic data (sex, age, and education), health information,
commonly based on screening tools [2e5], or individual nutrition including number of active diagnoses, the Charlson Comorbidity
parameters such as weight loss [7] which do not fully capture Index [16] (CCI), and number of medications, were also collected at
nutritional status; use of screening methods specifically which are admission. Length of stay in hospital and on each ward/unit
designed to have a high sensitivity, can inflate costs due to excess (medical, surgical, mixed medical/surgical, and ICU), and post-
prevalence [6]. While there is no gold standard for nutritional admission transfers, including dates of transfers, between units
assessment, the subjective global assessment (SGA) is a validated were documented for each participant. Patient charts were
nutritional assessment tool demonstrating high sensitivity and reviewed approximately every two days during their hospital stay
specificity and recommended as a predictive tool for clinical out- in order to track transfers and update health information as
comes [10e12]. Yet, relatively few reports have used the SGA to necessary. Hospital management and administrators completed a
assess for malnutrition [6,8,9]. standardized questionnaire that included average costs for medical,
Commonly, length of stay (LOS) is the focus for deriving costs surgical, and ICU beds.
with national or insurance company averages used to cost this care
and derive resource utilization estimates [2e4,7e9]. However, it is 2.1. Deriving costs
unclear if these administrative data for costing are consistent with
individual hospital data, which have rarely been used to date [5]. LOS in medical, surgical, mixed medical/surgical, and ICU beds
Further, not all studies adjust for the effect of co-morbidities on the was calculated for each participant by tracking the units patients
costs attributed to malnutrition [5,6,8]. For example, one study [9] were admitted to and transferred between. Bed days in each unit
controlled only for cancer and surgery during admission as markers were calculated by summing all bed days spent in that type of unit.
of morbidity in their analyses and another [2] made no The average cost of a medical, surgical, mixed medical/surgical, and
adjustments. ICU bed day was provided by each site. In addition, standardized
The Nutrition Care in Canadian Hospitals study [13] (NCCHS) data on hospital per diem expenses and direct expenses for func-
provides the opportunity to complete a cost analysis overcoming tional units were purchased from the Canadian Institute for Health
the deficits noted in prior work. Specifically, this was a multi-center Information (CIHI) (for comparison to site provided costs (i.e.,
cohort study of more than 1000 patients, where SGA was used at sensitivity analyses)).
admission to diagnose malnutrition. Patient information collected Three costs were derived for each patient in the study; total
also included demographic information, comorbidity and length of costs, medical costs and surgical costs. Total cost was the product of
stay and type of ward admitted and transferred to during their the number of bed days each patient spent on a each unit type
hospitalization. A site questionnaire provided specific hospital in- (medical, surgical, mixed, and ICU) and the average cost of a bed
formation including average daily costs for each type of ward. day on that unit type as recorded in a hospital questionnaire. For
Finally, costs recorded in a national data base, which holds stan- example, medical bed costs were calculated by multiplying the
dardized cost data for specific hospitals across Canada, is used as a number of days the patient occupied any medical bed during their
comparator for site reported costs. The goals of this study were to hospital stay by the average cost of a medical bed in that hospital.
determine: how malnutrition and hospital costs are associated; Surgical, mixed ward, and ICU costs were derived in a similar
whether adjusting for confounders changed the relationship; manner. All costs were calculated in 2012 real terms (adjusted for
whether the relationships were driven by medical or surgical pa- inflation). One hospital (n ¼ 57) classified all patient bed days as
tients or severity of malnutrition; and whether hospital derived mixed but provided costs for medical and surgical bed day. These
cost estimates provided similar information to a standardized na- observations were included in total costs by multiplying the
tional database. To our knowledge, the last two goals have not been number of mixed bed days by the average of the medical and sur-
previously studied. gical bed cost for that hospital. These observations were excluded
when deriving medical or surgical bed days and costs. One hospital
2. Methodology (n ¼ 59) provided costing information for medical wards only.
Surgical ward costs were assumed to be equal to medical ward costs
The Canadian Malnutrition Task Force conducted a prospective for the hospital. One hospital (n ¼ 40) provided all the data needed
cohort study [13], NCCHS, from July 2010 to February 2013 which for the study except for costs. To maintain the observations, and be
included patients 18 years or older, admitted to surgical or medical as conservative as possible, the medical and surgical costs for this
wards in one of 18 participating acute care hospitals that had vol- hospital were set equal to the minimum medical or surgical costs
unteered to participate. All provinces were represented except recorded in the data. Sensitivity analyses were performed around
Newfoundland and Prince Edward Island. Patients were excluded if all assumptions and with costing data purchased from CIHI (see
they were not able to give consent or were admitted directly to online appendix).
intensive care (ICU), obstetric, psychiatry, palliative care or medical
day units. Patients were enrolled according to a strict protocol to 2.2. Nutrition and covariates
avoid selection bias [14]. Days of enrollment rotated from Monday
to Friday, with Monday capturing the week-end admissions. Patients with all categories of SGA, well nourished (comparator),
Consecutive admissions were approached for consent and a moderately and severely malnourished were included in analyses.
maximum of 7 patients were followed at the same time to promote Previous studies indicate that malnourished patients tend to be:
feasibility. The study was approved by all institutions' administra- older, more likely to be male, and more likely to suffer from
tion and research ethic's boards (REBs) and all participants or their comorbidities [4,8]. Province of residence has been found to be a
alternative decision maker, where allowed by the hospital REB, strong predictor of health status, probability of hospital admission,
signed a consent form. The final study sample was 1022 patients and/or length of hospital stay [17,18] in Canada, and academic
recruited from community and academic hospitals. hospitals tend to have higher costs than community hospitals [19].
Patient's nutritional status was assessed at admission. SGA [15] These variables, which may affect health and/or hospital LOS and
was performed by 18 trained coordinators (one in each hospital) to costs, regardless of nutritional status, are added as controls when
avoid inter-rater variability (SGA A ¼ well-nourished; SGA examining the conditional associations between malnutrition and
B ¼ moderately malnourished; SGA C ¼ severely malnourished). hospital stay and/or hospital costs. Three proxies for health status
 L.J. Curtis et al. / Clinical Nutrition 36 (2017) 1391e1396 1393

were included initially; the CCI [16], the number of diagnoses, and Table 1
the median number of medications prescribed during admission Survey population descriptive statistics.

[14]. CCI was missing for 12% of the sample and never approached Variable Main samplea (n ¼ 958)
statistical significance in initial regressions, so it was dropped from [Standard errorb]
the analyses. Diagnoses were categorized as one, two or three or
Well nourish 55.7% [3.25]
more. Median number of medications is a continuous variable
Mod malnourish 33.5% [3.32]
ranging from zero to 35.5 (one patient had zero medications). Age, Sev malnourish 10.8% [1.58]
gender, socioeconomic status, living arrangement, type of hospital, Male 51.5% [2.46]
province and year were also added as controls. Age at time of Age 64.05 [0.85]
<High school 21.3% [3.20]
survey is a continuous variable and gender categorical (male ¼ 1,
High school 38.1% [2.08]
female ¼ 0). Education, classified as less than high school, >High school 40.6% [3.57]
completion of high school, secondary school completion, and post- Live with other 65.1% [2.89]
secondary education (comparator) proxied for socioeconomic sta- Live alone 26.5% [2.78]
tus. Multiple living arrangements were recorded in the survey. Residential 9.0% [1.79]
Number meds 12.64 [0.64]
These were categorized as living alone, living in residential care or
One diagnosis 64.0% [3.78]
living with someone in the community (comparator). Two diagnoses 26.2% [2.34]
3þ diagnoses 9.8% [1.71]
3. Analysis Academic hosp 65.9% [11.3]
Total beddays 9.86 [0.71]
Medical days 9.97 [0.95]
Stata/MP 13.1 [20] was used for all statistical analysis. As LOS Surgical days 7.92 [0.50]
and corresponding hospital costs were severely skewed and re- Total costs $ 6345 [649]
siduals were not normally distributed, linear regression with log- Medical costs $ 6214 [726]
arithmic transformation of the dependent variables [21] was Surgical costs $ 5355 [975]

undertaken. This type of analysis was possible due to the absence of a

Percent (%) of sample for binary variables and sample mean for
zero values and correct retransformation [22e24]. As observations continuous variables.
b
Standard Errors adjusted for clustering.
were clustered at the hospital level, all standard errors were
Source: Author's calculations.
adjusted for non-independence of observations. Sixty-four obser-
vations were excluded from the original study sample (n ¼ 1022)
due to missing information in covariates. In hospital mortality malnourished. Males make up almost 52% of the sample and the
resulted in censoring of LOS for 22 patients (2 were well nourished mean age is 64 years. About one fifth of the sample did not com-
(0.4% of well-nourished patients), 14 were moderately malnour- plete high school, 38% did and just over 40% obtained education
ished (4.5% of malnourished patients) and 6 were severely beyond high school. Over two-thirds of the sample lived with other
malnourished (5.9% of severely malnourished patients) (differences people before being admitted to hospital, one quarter lived alone,
are significant across groups)). The patients were included in the and almost one-tenth lived in a residential setting. The median
study with LOS as recorded at time of death as nutritional status is number of prescribed medications was, on average, close to 13. Just
strongly associated with mortality in these data (see online ap- over two-thirds of participants had one diagnosis while 26% had
pendix). The odds a moderately (severely) malnourished patient two diagnoses and almost 10% had three or more diagnoses. Just
died were 11 (13) times more than the odds a well-nourished pa- over two-thirds were admitted to academic hospitals.
tient did within the study period. Other studies have also shown a On average, patients spent almost 10 days (1e117 days) in
correlation between malnutrition and mortality [9]. The final esti- hospital leading to a total average cost of $6345 ($550e$89,498). A
mating sample size is 958. medical bed stay was almost 10 days (1e117 days), on average,
To illuminate the association between malnutrition and hospital costing an average $6214 ($347e$53,536) while a surgical stay was
costs, unadjusted analyses and then analyses adjusted for key 8 days (1e93 days) costing $5355 ($582e$49,222).
covariates are presented. Sensitivity analyses (available in online Mean outcomes by nutritional status are presented in Table 2.
appendix) were performed and indicate inclusion/exclusions and The average number of days spent in hospital was just over 3.2 (3.3)
assumptions do not drive study conclusions.
The coefficients, particularly on dummy variables, from the
regression model, lnðYÞ ¼ a þ b  N þ c  X þ e (N repre- Table 2
sents nutritional status, X represents control variables, and e are Outcome mean by nutritional status.
errors) are not as straightforward as often interpreted. Herein the Outcome Moderately Severely
regression results are presented as marginal effects. The marginal
Nourished Malnourished Malnourished
effects provide the correct interpretation of the change (propor-
[Standard errora]
tional (or percentage change if multiplied by 100)) in Y (not ln(Y))
for a small change in X (dummy variable changes from 0 to 1 or Survey data
continuous variables are increased by 1) [22,24]. As previously Beddays
Total (n ¼ 958) 8.43 [0.65] 11.66 [1.17] 11.70 [0.96]þ
stated, p-values are adjusted for clustering. Finally, given logarith- Medical (n ¼ 632) 8.20 [0.95] 12.05 [1.32]þ 12.05 [1.33]þ
mic transformation of the dependent variable, the appropriate Surgical (n ¼ 301) 6.98 [0.65] 9.62 [1.39]ǂ 8.75 [1.11]
adjusted predicted mean is reported for each regression analysis Costs
[25e27]. Total (n ¼ 958) $ 5074 [512] $ 7931 [766]þ $ 7989 [976]þ
Medical (n ¼ 632) $ 4839 [593] $ 7825 [849]þ $ 7823 [1042]þ
Surgical (n ¼ 301) $ 4303 [681] $ 7154 [1660] $ 6744 [1435]
4. Results
*, þ, ǂ indicates significantly different from well-nourished at 0.00, 0.05, 0.10,
respectively. Severely malnourished is never significantly different than moderately
Table 1 presents descriptive statistics for the full sample. Slightly malnourished.
over half of the participants are well nourished while one third are a
Standard errors are adjusted for clustering at hospital level.
moderately malnourished and just over one tenth are severely Source: Author's calculations.
1394 L.J. Curtis et al. / Clinical Nutrition 36 (2017) 1391e1396

days longer for a moderately (severely) malnourished patient, Table 4

when compared to a well-nourished patient (8.4, 11.7, and 11.7, Loglinear regression results for bed days.

respectively; p < 0.05 indicates significant difference between well- Total Medical Surgical
nourished and malnourished, difference between moderately and (n ¼ 958) (n ¼ 632) (n ¼ 301)
severely malnourished is n.s.). The association between medical
Marginal Marginal Marginal
bed days and nutritional status was similar with moderately and
severely malnourished patients spending significantly longer in Effect [p-value]* Effect [p-value]* Effect [p-value]*
medical beds, conditional on having stayed in a medical bed, (3.9 Mod malnourish 0.184 [0.014]þ 0.229 [0.014]þ 0.316 [0.015]þ
days; p < 0.05) than well-nourished patients. The trend was similar Sev malnourish 0.342 [0.000]* 0.527 [0.001]þ 0.167 [0.236]
for surgical stays but the difference was only barely statistically Male 0.001 [0.492] 0.062 [0.122] 0.019 [0.423]
Age 0.005 [0.014]þ 0.006 [0.009]þ 0.004 [0.048]þ
significant for moderately malnourished patients. Total cost <High school 0.062 [0.245] 0.060 [0.250] 0.056 [0.299]
increased by $2857 when patients were moderately malnourished High school 0.002 [0.475] 0.100 [0.018]þ 0.035 [0.328]
and by $2915 when severely malnourished (~60%, p < 0.05). For Live alone 0.013 [0.376] 0.074 [0.137] 0.042 [0.333]
medical bed stays, costs were significantly higher (p < 0.05), $2986 Residential 0.253 [0.010]þ 0.253 [0.005]þ 0.237 [0.188]
Number meds 0.019 [0.002]þ 0.013 [0.011]þ 0.013 [0.068]
for moderate and $2984 for severely malnourished patients. The
Two diagnoses 0.229 [0.007]þ 0.268 [0.006]þ 0.093 [0.235]
average cost difference between a well-nourished and moderately 3þ diagnoses 0.568 [0.000]* 0.548 [0.002]þ 0.536 [0.102]
malnourished surgical patient was $2851 and $2441 for severely Academic hosp 0.116 [0.065]ǂ 0.053 [0.300] 0.046 [0.324]
malnourished patients but the differences were not statistically Predicted mean 9.407 9.537 7.768
significant. R-squareda 0.1860 0.2046 0.1778
Univariate analyses on the transformed outcomes are presented
*p-value adjusted for clustering.
in Table 3 and multivariate analyses with added controls are pre- *, þ, ǂ indicates significantly different from well-nourished at 0.00, 0.05, 0.10,
sented in Tables 4 and 5. The results shown in Table 3 are the respectively.
a
marginal effects of malnutrition, relative to a well-nourished pa- Results control for province and year.
tient, with no adjustment for covariates, the standard errors have Source: Author's calculations.

been adjusted for clustering at hospital level. The predicted mean

LOS (9.3 days) is very close to the sample mean provided in Table 1.
Table 5
Results indicate that nutritional status is strongly and significantly Loglinear regression results for costs.
associated with LOS. Moderate malnourishment is associated with
Total costs Medical costs Surgical costs
a 27% increase (p ¼ 0.006), and severe malnutrition with a 44%
increase (p ¼ 0.000), in total hospital days. Length of stay increases (n ¼ 958) (n ¼ 632) (n ¼ 301)
substantially more, on average, for medical stays with a 40% in- Marginal Marginal Marginal
crease for moderate malnutrition (p ¼ 0.008) and a 59% (p ¼ 0.002) Effect [p-value]* Effect [p-value]* Effect [p-value]*
increase for severely malnourished patients. Significance wanes for
Mod malnourish 0.312 [0.003]þ 0.285 [0.004]þ 0.336 [0.022]þ
surgical stays (potentially due to small sample size); malnutrition is
Sev malnourish 0.378 [0.002]þ 0.550 [0.003]þ 0.269 [0.145]
associated with 28% (p ¼ 0.014) and 29% (p ¼ 0.067) increase for Male 0.012 [0.412] 0.024 [0.338] 0.066 [0.243]
moderately and severely malnourished patients, respectively. Age 0.005 [0.011]þ 0.006 [0.009]þ 0.007 [0.011]þ
Moderate malnutrition is associated with a 45% (p ¼ 0.002) in- <High school 0.022 [0.405] 0.149 [0.065] 0.013 [0.460]
crease in total costs, a 55% (p ¼ 0.004) and 52% (p ¼ 0.062) increase, High school 0.024 [0.384] 0.087 [0.034]þ 0.013 [0.431]
Live alone 0.059 [0.067] 0.035 [0.299] 0.055 [0.434]
respectively, for medical and surgical costs. Severe malnutrition is Residential 0.263 [0.067]ǂ 0.288 [0.007]þ 0.105 [0.311]
associated with a 54% (p ¼ 0.006), a 68% (p ¼ 0.009) and a 51% Number meds 0.012 [0.031] 0.007 [0.085] 0.004 [0.290]
(0.066) increase, respectively, for total, medical, and surgical costs. Two diagnoses 0.134 [0.035]þ 0.168 [0.028]þ 0.051 [0.334]
3þ diagnoses 0.435 [0.002]þ 0.400 [0.006]þ 0.438 [0.115]
Academic hospital 0.360 [0.100]ǂ 0.198 [0.104] 1.144 [0.001]þ
Predicted mean 6195 6219 5208

R-squareda 0.2930 0.3375 0.3409

Table 3
Loglinear regression univariate results. *p-value adjusted for clustering.
*, þ, ǂ indicates significantly different from well-nourished at 0.00, 0.05, 0.10,
Total Medical Surgical respectively.
a
Results control for province and year.
(n ¼ 958) (n ¼ 632) (n ¼ 301)
Source: Author's calculations.
Marginal Marginal Marginal

Effect [p-value]* Effect [p-value]* Effect [p-value]*

R2s are low for all regressions but this is not unusual for health
Bed days
utilization studies [17].
Mod malnourish 0.272 [0.006]þ 0.399 [0.008]þ 0.284 [0.014]þ
Sev malnourish 0.438 [0.000]* 0.594 [0.002]þ 0.294 [0.067]ǂ Table 4 demonstrates that nutritional status remains signifi-
Predicted mean 9.344 9.251 7.668 cantly associated with LOS after adjusting for individual and hos-
R-squared 0.0346 0.0549 0.0353
pital characteristics, which may drive LOS independently of
nutritional status. Moderate malnourishment is independently
Costs
associated with an average increase of 18% in hospital days
Mod malnourish 0.450 [0.002]þ 0.553 [0.004]þ 0.524 [0.062]ǂ
Sev malnourish 0.543 [0.006]þ 0.678 [0.009]þ 0.514 [0.066]ǂ (p ¼ 0.014), 23% increase in medical days (p ¼ 0.014), and 32% in-
Predicted mean 6157 6030 5164 crease in surgical stays (p ¼ 0.015) when compared to a well-
R-squared 0.0475 0.0628 0.0682
nourished patient. Severely malnourished patients stay 34%
longer (p ¼ 0.000) in hospital and approximately 53% longer
*p-value adjusted for clustering at hospital level.
(p ¼ 0.001) in medical beds than do well-nourished patients with
*, þ, ǂ indicates significantly different from well-nourished at 0.00, 0.05, 0.10,
respectively. similar characteristics. Adding covariates eliminates the signifi-
Source: Author's calculations. cance of the relationship between severe malnutrition and surgical
 L.J. Curtis et al. / Clinical Nutrition 36 (2017) 1391e1396 1395

bed days (17% longer but p ¼ 0.236). Patient covariates indepen- which translates to about $1500 to $2000 (2012 dollars) after
dently associated with longer LOS are: increasing age, living in controlling for characteristics that may influence the cost of hos-
residential settings, more medications prescribed, and having more pitalization and the likelihood of being malnourished. The signifi-
diagnoses during their hospital stay. Education is not significantly cance in the relationship wanes when examining surgical stays but
related to LOS except that high school graduates have, on average, this may be due to the small sample size for surgical bed stays or
fewer medical bed days than do those with post-secondary edu- the possibility that some severely malnourished patients may not
cations (p ¼ 0.018). The only hospital characteristics independently be candidates for some surgical procedures.
associated with longer LOS are academic (p < 0.10) and province The results of this study are consistent with prior work [3,4,9]
(results not shown). however what is unique about this study is the adjustment of
Cost results are presented in Table 5. Adjusting for covariates diverse patient and hospital characteristics as well as the clustering
diminishes the relationship slightly but the independent effect of effect of hospital. Prior work has not adjusted for as an extensive set
malnutrition on hospital costs remains large in magnitude and of covariates or recognized that LOS and costs are dependent on the
strongly significant. Moderately malnourished patients cost, on individual hospital setting. It is striking that the effect of malnu-
average, between 31% and 34% more than well-nourished patients trition is more important than any covariate, excepting age and the
(p-values < 0.05). Severe malnourishment is associated with higher number of diagnoses which are also strong drivers of hospital costs.
total and medical costs, on average; 38% higher total costs In addition, most work to date does not differentiate on the
(p ¼ 0.003), 55% higher medical costs (p ¼ 0.003), but surgical costs severity of malnutrition. Prior work has demonstrated that the
are not significantly higher for malnourished patients. The only moderately malnourished group is not readily identified in referral
patient covariates independently associated with costs were procedures in Canadian hospitals [28]. Current results indicate that
increasing age, number of medications, and number of diagnoses. moderately and severely malnourished patients have similar out-
Academic hospitals were associated with higher hospital costs and comes and, as such, all malnourished patients should be identified
particularly higher surgical costs. on admission so that appropriate nutritional care can be instituted
Sensitivity analyses (presented in online appendix) were as early as possible.
completed to test changes in samples (excluding one Quebec hos- Sensitivity analyses determined that the independent effect of
pital recording only ‘mixed beds’ (‘limited sample’ n ¼ 900) and malnutrition on hospital costs derived from LOS holds while using a
dropping all Quebec hospitals (‘No Quebec’ n ¼ 787)), assumptions, variety of costing information providing a notion of generalizability
and costing data. across studies with different costing data. Future studies should
Online Appendix Table 1 documents mean outcomes for add to this study by focusing on counting patient level costs
different samples and Table 2 shows outcomes by nutritional sta- incurred by malnourished as compared to well-nourished patients
tus. Results are very similar to Tables 1 and 2 of the main analyses. during their hospital stay (e.g., added costs of nutritional supple-
The only differences worthy of note are the average costs ments, nutritionist and other therapists visits, nutrition related
depending on source of costing data. Average costs calculated with pharmaceutical costs, etc.) rather than using aggregate costs per
CIHI Direct Expenditure data were close to $1000 lower than bed day.
average costs using comparable data provided by the hospitals Finally, analyzing medical and surgical patients separately pro-
while average costs calculated using CIHI Per Diem costs are almost vided the opportunity to identify differences among the two groups
double the average costs using the comparable hospital data. This of patients. The association between malnutrition and LOS/costs
demonstrates that costing definitions/inclusions differ depending waned for surgical patients, especially the severely malnourished.
on the data source and sensitivity analyses are required to assure However, it should be noted that sample size was small and, in
results are generalizable. some cases, patients were transferred between units. There were 14
Online Appendix Tables 3 and 4 present the regression co- patients who transferred between ICU and medical beds, 8 patients
efficients on our variables on interest (coefficients on control var- who transferred between ICU and surgical beds, 14 patients who
iables are not reported) to test sensitivity to sample specifications transferred between ICU and both medical and surgical beds, and
(Table 3) and costing assumptions (Table 4). Although significance 19 patients who transferred between medical and surgical beds.
between severe malnutrition and surgical costs are insignificant in While this study is the first, that we are aware of, to point out the
most cases (similar to our full-sample results), the relationships differences between medical and surgical stays, the small sample of
between moderate and severe malnourishment hold strong in surgical patients and the extremely small samples of transfer pa-
alternate samples and for alternate costing data sources. The tients (disallowing further sub-population analyses), make it diffi-
conclusion that malnourished patients' hospital stays cost signifi- cult to draw any strong conclusions regarding differences across
cantly more than well-nourished patients with similar character- patients that inhabit specific types of hospital beds. Future research
istics holds. with larger sample sizes would be necessary to ascertain whether
the conclusions are generalizable.
5. Discussion Although this work overcame many of the weaknesses seen in
prior research, there are some limitations in additions to those
This study identified that almost half of hospital patient par- already discussed. As is typical in any cohort study, there were
ticipants in the Nutrition Care in Canadian Hospitals study were some missing data on covariates, which reduced the sample in
malnourished at admission as measured by the SGA; one third were analyses to 958 from the original participant pool of 1022. Multi-
moderately malnourished and just over one tenth were severely collinearity may also be an issue, particularly for surgical bed days,
malnourished. Malnourished patients stay in hospital longer given the marginal effects and R2 are similar to the other re-
(approximately three days longer, on average than well nourished) gressions but standard errors are substantially larger. We did not
and, as a result, cost substantially more, on average, per hospital test for potential interactions among variables used to adjust the
stay. The strong univariate relationship, adjusting for non-normal association between malnutrition and LOS/costs as this would have
distribution of residuals and clustering of observations, between increased the effects of multicolinearity.
malnourishment and hospital stays/costs held after adjusting for In conclusion, moderate and severe malnutrition are individu-
patient and hospital characteristics. Malnutrition at admission in- ally and independently associated with LOS and costs in Canadian
creases total costs by 31% (moderate) to 38% (severe), on average, hospitals. The results were particularly strong for patients who
1396 L.J. Curtis et al. / Clinical Nutrition 36 (2017) 1391e1396

spent at least one night in a medical bed. The results indicate that References
approximately 40% of patients were malnourished at admission
and they cost between $1500 and $2000 more per hospital stay [1] McKinlay AW. Malnutrition: the spectre at the feast. J R Coll Physicians Edinb
2008:38317e21.
than a well-nourished patient. Hospital discharge data that come [2] Alvarez-Herna ndez JM, Planas Vila M, Leo  n-Sanz M, García de Lorenzo A,
from CIHI [29] indicate that approximately 2.6 million adults were Celaya-Pe rez S, García-Lorda P, et al. Prevalence and costs of malnutrition in
discharged from Canadian hospitals in 2012/2013. A very rough hospitalized patients; the PREDyCES® Study. Nutr Hosp 2012;27(4):1049e59.
[3] Amaral TF, Matos LC, Tavares MM, Subtil A, Martins R, Nazare  M, et al. The
‘back of the envelope’ analysis using these figures indicates the economic impact of disease-related malnutrition at hospital admission. Clin
total additional cost to the health care system that may be attrib- Nut 2007;26:778e84.
uted to malnutrition at hospital admission is substantial at [4] Freijer K, Tan SS, Koopmanschap MA, Meijers JMM, Halfens RJG, Nuijten MJC.
The economic costs of disease related malnutrition. Clin Nutr 2013;32:
approximately $1.56 to $2.1 billion per year. This figure may only be 136e41.
a starting point as hospital costs may poorly capture the total [5] Gastalver-Martin C, Alarco -Payer C, Leo n-Sanz M. Individualized measure-
economic impact of malnutrition. The repercussion of malnutrition ment of disease-related malnutrition's costs. Clin Nutr 2015;34(5):951e5.
[6] Guerra RS, Sousa AS, Fonseca I, Picehl F, Restio MT, Ferreira S, et al.
on a patient's ability to return to work or to require additional
Comparative analysis of undernutrition screening and diagnostic tools as
outpatient care was not addressed. Future work needs to determine predictors of hospitalization costs. J Hum Nutr Diet 2016;29(2):165e73.
if improved care practices, as suggested in the Integrated Nutrition [7] Melchior JC, Preaud E, Heurtebis J, Brami M, Duru G, Fontaine E, et al. Clinical
and economic impact of malnutrition per se on the postoperative course of
Pathway for Acute Care [30] can divert any additional costs and
colorectal cancer patients. Clin Nutr 2012;31(6):896e902.
improve patient outcomes. [8] Lim SL, Ong KCB, Chan YH, Loke WC, Ferguson M, Daniels L. Malnutrition and
its impact on cost of hospitalization, length of stay, readmission, and 3-year
mortality. Clin Nutr 2012;31:345e50.
Funding source [9] Correia MI, Waitzberg D. The impact of malnutrition on morbidity, mortality,
length of hospital stay and costs evaluated through a multivariate model
analysis. Clin Nutr 2003;22(3):235e9.
The Canadian Nutrition Society (CNS) provided the funds for the [10] da Silva Fink Jaqueline, Daniel de Mello Paula, Daniel de Mello Elza. Subjective
conduct of this study. Unrestricted grants were received by CNS global assessment of nutritional status - a systematic review of the literature.
Clin Nutr 2015;34(5):785e92.
from Abbott Nutrition, Nestle Nutrition, Baxter, Fresenius Kabi, and
[11] NEMO Consensus document from Dietitian/Nutritionists from the Nutrition
Pfizer during the period of 2010e2013 for the conduct of this study. Education Materials Online “NEMO” team [Accessed 27 April 2016] https://
www.health.qld.gov.au/nutrition/resources/hphe_scrn_tools.pdf.
[12] van Bokhorst-de van der Schueren Marian, Realina Guaitoli Patricia,
Conflict of interest Jansma Elise P, de Vet Henrica CW. Nutrition screening tools: does one size fit
all? A systematic review of screening tools for the hospital setting. Clin Nutr
2014;33(1):39e58.
Dr. Curtis reports no conflicts of interest for this study. [13] Nutrition Care in Canadian Hospitals. Canadian Malnutrition Task Force.
Mrs. Bernier reports personal fees and other from Canadian [Accessed 18 June 2016] at http://nutritioncareincanada.ca/malnutrition/.
[14] Allard JP, Keller H, Jeejeebhoy KN, Laporte M, Duerksen DR, Gramlich L, et al.
nutrition society through an unrestricted grant from Abbott Malnutrition at hospital admission: contributors and impact on length of stay:
nutrition during the conduct of the study; personal fees from a prospective cohort study from the Canadian Malnutrition Task Force. JPEN J
Abbott nutrition outside the submitted work; and President, Ordre Parenter Enter Nutr 2016;40(4):487e97.
te
tistes du Que bec this is the licensing and [15] Detsky AS, McLaughlin JR, Baker JP, Johnston N, Whittaker S, Mendelson RA,
professionnel des die
et al. What is subjective global assessment of nutritional status? JPEN J
regulatory body for dietitians in Quebec Elected April 1 2014. Parenter Enter Nutr 1987;11(1):8e13.
Dr. Jeejeebhoy reports grants from Abbott Inc. during the [16] Charlson ME, Ales KA, Pomper P, MacKenzie CR. A new method of classifi-
cation of prognostic comorbidity for longitudinal studies development and
conduct of the study; other from Seaford Pharmaceutical, personal
validation. J Chronic Dis 1987;40(5):373e83.
fees from Robarts Clinical Trials, outside the submitted work. [17] Curtis LJ, MacMinn WJ. Health care utilization in Canada: twenty-five years of
Dr. Allard no conflicts of interest for this study. evidence. Can Public Policy 2008;34(1):65e88.
Dr. Duerksen reports personal fees from Abbott Nutrition, [18] Health Canada. Canada’s Health Care System (Medicare) [Accessed 11 August
2015] http://www.hc-sc.gc.ca/hcs-sss/medi-assur/index-eng.php.
outside the submitted work. [19] Ayanian John Z, Weissman Joel S. Teaching hospitals and quality of care: a
Dr. Gramlich reports no conflicts of interest for this study. review of the literature. Milbank Q 2002;80(3):569e93.
Dr. Laporte reports other from Support for travel to meetings for [20] StataCorp. Stata statistical software: release 13. College Station, TX: StataCorp
LP; 2013.
the Nutrition Care in Canadian Hospital Study design and data [21] Wen Chuck K, Chambers Catharine, Fang Dianne, Mazowita Garey,
analysis, during the conduct of the study. Hwang Stephen W. Length of stay and hospital costs among patients admitted
Dr. Keller reports grants from Canadian Nutrition Society, to hospital by family physicians. Can Fam Physician 2012;58:290e6.
[22] Kennedy PE. Estimation with correctly interpreted dummy variables in
outside the submitted work; and Abbott Nutrition, Nestle Health semilogarithmic equations. Am Econ Rev 1981;71(4):801.
Sciences have supported the Canadian Malnutrition Task Force of [23] Montez-Rath Maria, Christiansen Cindy L, Ettner Susan L, Loveland Susan,
the Canadian Nutrition Society for several years through educa- Rosen Amy K. Performance of statistical models to predict mental health and
substance abuse cost. BMC Med Res Methodol 2006; Oct 26;6:53.
tional grants and a member of the Speaker's Bureau for both or-
[24] van Garderen K, Shah C. Exact interpretation of dummy variables in semi-
ganizations in the past two years. logarithmic equations. Econom J 2002;5(1):149e59.
[25] Cameron AC, Trivedi P. Microeconometrics using stata. Stata Press; 2009.
[26] Duan N. Smearing estimate: a nonparametric retransformation method. J Am
Acknowledgments Stat Assoc 1983;78:605e10.
[27] Nichols A. Regression for nonnegative skewed dependent variables. BOS'10
Stata Conference. [Accessed 18 June 2016] at http://repec.org/bost10/nichols_
LC designed the methodology, amalgamated and cleaned the boston2010.pdf.
data, carried out the analyses, and was the primary writer of the [28] Dieticians of Canada - Ontario Clinical Nutrition Leaders Action Group. An
Inter-professional Approach to Malnutrition in Hospitalized Adults: Dietitians
original and revised papers. HK worked in close collaboration with Leading the Way. Dieticians of Canada, June 2014. [Accessed 18 June 2016] at
LC during analysis and writing. KJ and PB provided conceptualiza- www.dietitians.ca.
tion of the analysis during early stages and reviewed a draft of the [29] Canadian Institutes for Health Information: Quick Stats [Accessed 18 June
2016] at https://www.cihi.ca/en/quick-stats.
paper. JA, DD, LG, ML coordinated data collection at their sites,
[30] Keller H, McCullough J, Davidson B, Vesnaver E, Laporte M, Gramlich L, et al.
provided guidance to other sites for data collection, provided some The Integrated Nutrition Pathway for Acute Care (INPAC): building consensus
guidance on direction of the paper and reviewed a draft of the with a modified Delphi. Nutr J 2015;14:63. http://dx.doi.org/10.1186/s12937-
015-0051-y. published online June 19, 2015.
paper.