The following pages constitute the final, accepted and
revised manuscript of the article:
Borg, Sixten and Persson, Ulf and Odegaard, Knut and Berglund, Göran and
Nilsson, Jan-Ake and Nilsson, Peter M
“Obesity, survival, and hospital costs-findings from a
screening project in sweden.”
Value Health. 2005 Sep-Oct;8(5):562-71.
Publisher: Blackwell-Synergy.
Use of alternative location to go to the published version
of the article requires journal subscription.
Alternative location: http://dx.doi.org/10.1111/j.1524-4733.2005.00048.x
1
2005-02-08/ SB,UP,KÖ
Original paper
For Value in Health
Obesity, survival and hospital costs findings from a screening project in Sweden
Sixten Borg1, Ulf Persson1, Knut Ödegaard1, Göran Berglund2 , Jan-Åke Nilsson2 and Peter M Nilsson2,
1. IHE (The Swedish Institute for Health Economics) Box 2127, S-22002 Lund, Sweden
2. Department of medicine, University hospital, Malmö, S-205 02 Sweden
Short title:
obesity, survival and hospital costs
Number of words: 3930 excluding reference list and tables and 262 words in abstract .
Correspondence: Ulf Persson, IHE Phone: +46-46-32 91 00 Fax: +46-46-12 16 04
e-mail:
up@ihe.se
Alternative proof reader: Peter Nilsson, MD, PhD Phone: +46-40-33 24 15
Fax: +46-40-92 32 72
e-mail:
Peter.Nilsson@medforsk.mas.lu.se
Acknowledgements
We thank Pharmaceuticals Group Strategic Marketing, Johnson & Johnson, New Jersey, USA, for an
unrestricted research grant making this study possible.
2
Abstract
Objective: Our aim was (1) to estimate the costs of hospital treatment and the (2) value of lost production due to
early death associated with overweight and obese patients, and then to extrapolate the findings to national costs.
Methods: We use regression models to analyze survival, expected number of days in hospital treatment for
patients with different BMI and costs with data obtained from screening of 33196 middle-aged subjects living in
Malmö, Sweden, and collected during a 15-year follow-up period. We subsequently scale up costs to national
aggregate level using the BMI prevalence data from the screening project to the national population.
Results: The total excess hospital (somatic, psychiatric) care cost (SEK) for the national health care budget,
excess as compared to normal weight patients for obese (BMI>30) and overweight weight (25<= BMI < 30) was
estimated to SEK 2 155 million per annum (US$269 million, assuming $1=SEK8), or about 2.3 percent of total
hospital care costs in Sweden. The corresponding indirect costs due to early death were estimated to SEK 2 935
million (US$367 million). For males at age 55 the potential hospital costs saving, excluding costs of the
intervention, that could be gained by an intervention that successfully and safely could alter the weight of an
obese individual to become normal weight was estimated to on average SEK 4 434 (US$ 554) per annum.
Conclusion: Hospital treatment costs are found to be higher for obese and overweight patients than for normal
weight patients indicating potential cost savings especially on indirect costs by effective, safe and low cost
weight-loss intervention.
Keywords: BMI, obesity, hospital bed-days, survival, costs
Abstract number of words: 262
3
Introduction
Obesity is a well-known risk factor for increased morbidity and premature mortality [1-4]. WHO uses the body
mass index (BMI) as a criterion for defining obesity. This index is calculated as the ratio between a persons
weigh in kg and the persons' length in meter squared (kg/m2). When this ratio is greater than 30 a person is
defined to be obese and when the ratio take a value between 25 - 30 a person is defined to be overweight.
Normal weight are defined for persons with BMI 18.5 - 25 and subjects with BMI< 18.5 as underweight [5]. In
this study we will analyze hospitalization for subjects with BMI > 18.5 only.
Numerous observational studies from many populations, both in men and women, have found obesity to be
associated with cardiovascular diseases, type 2 diabetes and some cancer forms, e.g. breast cancer and
endometrial cancer of corpus uteri [6-9]. As a consequence of the association between obesity and increased
morbidity, longer hospital stays and greater risk for complications following medical intervention, obesity is also
associated with increased costs of treatment [10-16].
In the literature there are a few studies that have reported on the association between BMI and health care costs
[17-19]. All these studies provide estimates of the cost of illness (COI) attributed to obesity at one point in time
using the prevalence approach. The estimates are based on calculations of the total annual expenditure of a
particular disease, e.g. type 2 diabetes, multiplied by an obesity-etiological fraction that measures the impact of
obesity as a risk factor for type 2 diabetes. However, COI estimates based on the prevalence approach are not
very useful for guiding decisions on prevention because the time dimension of the illness development is not
considered. In order to provide information for decisions on prevention, the economic analysis needs to account
for the time between the investments in preventing efforts and the benefits in reduced treatment costs, i.e., the
time dimension. Cohort studies using the incidence approach can provide such observational information [20].
In this study we will seek to provide further evidence of the association between obesity and treatment costs. In
particular we seek to address two specific questions: (1) are the hospital treatment costs for obese and
overweight patients higher than for normal-weight patients and if so how much does this higher cost imply for
the national Swedish health care budget, (2) are there any differences in life expectancy between obese and
overweight patients from normal weight patients and if so how much is the associated costs for lost production.
4
Subjects and methods
Subjects
In order to answer the three questions posed, we used hospital data from medical records for middle-aged
subjects of both sexes, recruited from the Malmö Prevention Project (MPP). All subjects in MPP were originally
recruited from Malmö city, Sweden. Between 1974 and 1984, in all 22,444 men born between 1921 and 1949
(age range 35-51 years) and 10,533 corresponding women, constituting 70-75% of the total population in these
birth cohorts (85% born in Sweden, 99% Caucasian), took part in the MPP. The project was carried out at the
Department of Medicine, Malmö University Hospital, in southern Sweden. The aim of MPP was to screen the
local population for cardiovascular risk factors and alcohol over-consumption. The subjects had a mean followup of 17 years (range 0-24 years). We have limited this study to the first 15 years of follow-up, since the
proportion of patients lost to follow-up gradually accelerates beyond 15 years. Beyond 15 years we have too few
observation to base our analysis on. A total of 23 365 subjects (70%) had a follow-up period of at least 15 years.
Follow-up data were derived from register linkage analyses, covering a time period with expansion of the health
care sector and medical treatment modalities. However, due to financial restrictions in health care in general, the
mean number of days of each episode of hospitalization has gradually decreased for most common diagnostic
categories. The loss of subjects to follow-up is related at least partly to mortality in about 1/3 of these subjects,
as well as loss of a number of subjects that had emigrated abroad during follow-up.
Data on screening routines have previously been published as well as some follow-up analyses [21 - 24]. In
Malmö city, there is only one central university hospital for the local population (250 000 inhabitants). The
patients were classified into BMI Groups using the WHO criterion. Throughout this paper, BMI refers to
baseline BMI. Baseline characteristics of the subjects are presented in Table 1.
Although the study was designed as a screening project for cardiovascular events we believe the BMI data
generated from the study can also be used to answer the three questions that we have posed for this study. This
especially so since the Swedish system of using a 10-digit personal number for all citizens, making it possible to
trace persons even if they move around and will receive in-hospital care in different cities. Not many countries
have the same administrative system and therefore we believe a study on the long-term economic consequences
of obesity is suitable to do in Sweden.
5
To answer the questions that we have posed for this study we will estimate the following equation:
HAggregated = f(age, sex, BMI group)
(1)
where HAggregated = number of days in hospital over the follow-up time, which depends on age, sex and baseline
BMI group, i. e. normal weight, overweight and obese.
The cost of hospital treatment we subsequently estimate by multiply HAggregated with an average cost per day
of care in hospital. We use administrative prices for the costs of 24 hours hospital stay at the general medical
ward and the psychiatric ward from Malmö City University Hospital. These prices were for the year 2003 SEK
3899 at the medical ward and SEK 3055 at the psychiatric ward. We use a weighted average cost of inpatient
stay with weight 2/3 for the general medical ward and 1/3 for the psychiatric ward. These weights reflect roughly
the pattern of admission of the screened population.
Method and Statistical analysis
The observed data on hospitalization had a very skewed distribution. There was a tendency towards a pattern of
either no hospitalization at all in a given year, or frequent and in some cases lengthy hospitalization. Therefore
we tried to discriminate between two subgroups without hospitalisation and with hospitalisation within a given
year, and analyze the hospitalization rate in the latter one.
To obtain numerical estimate of equation (1) we use three regression equations, one for the survival time (T)
(equation 2), one for the probability (Pr) of hospitalization in a given year (equation 3), and one for the number
of days in hospital (H) given hospitalization (equation 4), defined as follows:
ln (T(D1, D2, D3, Age)) = α1 + α2*D1 +α3*D2 + α4*D3 + α5*Age + ε1
(2)
Pr {H(D1, D2, D3, Age) > 0} = (1 + exp(β1 + β2*D1 + β3*D2 + β4*D3 + β5*Age + ε2)) -1
(3)
y(D1, D2, D3, Age) = ln (H(D1, D2, D3, Age)) = (γ 1 + γ2 *D1 + γ3 *D2 +γ4*D3 + γ4*Age + ε3)
(4)
6
For equation (2) we use an accelerated failure time model where age, two dummy variables for BMI group (D1
and D2) and one dummy variable for gender (D3) are explanatory variables, and the survival times (T) are
assumed to have a Weibull distribution. This method assumes a parametric shape of the survival function,
whereas non-parametric methods such as the Kaplan-Meier estimator make no such assumptions. Our approach
enables the use of age as a continuous explanatory variable. The fitted parametric survival function is then used
to compute the probability of survival beyond t years of follow-up time, Pr{T(D1, D2, D3, Age) > t}. In equation
(3) we do a logistic regression of the probability of non-zero number of days in hospital in a given year. Finally
in equation (4), we do a linear regression of the log number of days in hospital conditional on non-zero number
of days. All these regression equations have age, BMI group and gender as explanatory variables. The dummy
variables D1, D2 and D3 is set equal to 1 for overweight, obese and male patients respectively otherwise 0, and
we divided the data into individual patient-years, so that each subject provided up to 15 years of observation.
The log-transform in equation (4) necessitates the use of a smearing estimator to avoid bias when using the
model to predict onto the natural scale [25].
Having obtained numerical estimates of equation (2), (3) and (4) numerical estimate of equation (1) is derived as
follows:
HAggregated(D1, D2, D3, Age) =
15
∑ {Pr{ T(D1, D2, D3, Age) > t}
* Pr {H(D1, D2, D3, Age+t-1) > 0} * exp(y(D1, D2, D3, Age+t-1)+s)}
t =1
where s is a smearing estimate for loglinear models, which is used to adjust for transformation bias and age in
equation 3 and 4, is substituted with age + t - 1 to account for follow-up time.
Indirect costs we define equal to the value of production that is lost due to death before retirement age. This
value we define equal to the number of years between age at death and age 65 times the annual wage including
the social costs of labor (social insurance, pension etc.) of the deceased. We follow standard convention and
assume that the wage reflect the value of production. We have assumed that all subjects in all BMI groups have
an average wage equal to the national average by age and gender [29].
7
Both healthcare costs and indirect costs due to loss of production were estimated for a period of 15 years, in male
and female subjects of ages 30-60, in each BMI group. We sum up the discounted costs over 15 years using a
discount rate of 3%. The differences as compared to normal weights were estimated on a yearly basis.
We make a final aggregation of these costs by projecting the cost estimates onto the Swedish population. Each
stratum in the population, defined by age (30-60), BMI group and gender, were multiplied by the predicted
excess cost weighted by relative prevalence within the study subjects. In this context, the normal weight subjects
yield zero costs in analogy with the methods described above. The resulting estimate is therefore the estimated
cost due to obesity and overweight as compared to the hypothetical situation where all are normal weight.
The R language was used for statistical analyses and graphics [26].
Results
Survival pattern
Our parameter estimates of equation (22) are reported in Table 2.
Our model for survival showed that male subjects have a higher risk of death than female subjects (p<0.0001),
and that the risk is higher in obese (p<0.0001) as compared to normal weight subjects. No difference was
detected between overweight and normal weight subjects. In addition, the model confirms the expected result
that the risk of death increases with age (p<0.0001). Predicted survival curves for 60-year old male in each BMI
group are shown in Figure 1.
Hospitalization rate.
Parameter estimates of equation (33) and (44) are reported in Tables 3 and 4.
Table 4 Parameter estimates probability of having non-zero hospitalization rate, equation (3)
Estimate Std. Error z value Pr(>|z|)
(Intercept) -4.1119733
0.0392840 -104.67
<2e-16 ***
age
0.0303135
0.0006731
45.03
<2e-16 ***
bmi.groupOB
0.4188961
0.0181166
23.12
<2e-16 ***
bmi.groupOW
0.1281724
0.0112593
11.38
<2e-16 ***
8
bmi.groupUW
0.3993402
0.0353681
11.29
<2e-16 ***
sexM
0.1650606
0.0117388
14.06
<2e-16 ***
The logistic model, equation (33), shows the risk on non-zero hospitalization to increase with age (OR 1.031,
95% C.I. 1.029-1.032 for one year older), for male as compared to female subjects (OR 1.18, CI 1.15-1.21), for
obese (OR 1.52, CI 1.47-1.58) and overweight (OR 1.14, CI 1.11-1.16) as compared to normal weight. The
predicted probabilities of hospitalization are shown in Figure 2, for 60-year old male subjects in each BMI
group.
Table 4 reports the parameter estimates for equation (44). The number of days in hospital increases by age
(p<0.0001), but no statistical significant difference in hospitalization by gender. The rate was significantly higher
for obese (p=0.0024) and significantly lower for overweight subjects (p=0.0026), as compared to normal
subjects. Unfortunately, the predictive power of the equation is rather modest, adjusted R2 about 0. 9 %.
However, in this study we are not seeking to predict the individual variation in number of days in hospital but
rather the difference between groups of individuals. Our model clearly shows that there are differences in the
number of days in hospital between groups of individuals with different BMI values and hence we will use this
model to answer the questions that we have posed for this study. Figure 3 shows predicted number of days in
hospital using equation 4 in male 60-year old subjects in each BMI group.
Expected incremental costs of hospitalization and indirect cost due to loss of production over 15 years
Using the results from equation 2 - 4 we can now estimate equation (1). The result is presented in Table 5. In this
table we report the expected numbers of days in hospital for age cohorts 30-60, accumulated over 15 years along
with the incremental cost of hospitalization as compared to normal weight subjects. A graphical presentation of
the expected number of days in hospital for male subjects of different BMI groups is given in Figure 4. The
curves in Figure 4 are generated as the product of the predicted probability of survival, the predicted probability
of hospitalization and the predicted number of days in hospital. Note that the curve for overweight lies above the
curves for normal weight, whereas in Figure 3 the curve for over weight lies below the curves for normal weight.
The reason why the curve for over weight shifts its position in Figure 4 is that over weight patients have a higher
probability of hospitalization.
9
The mean number of days in hospital of normal weight patients and overweight patients are, however, very
similar, while obese patients have more days than normal weight. In Table 5 one may also note that the number
of days in hospital increases by age in all BMI groups.
Estimates of the indirect costs due to loss of production are reported in Table 6. Overweight and normal weight
being very similar, while obese have greater loss. The indirect cost of lost production increases by age up to age
50 because average annual salary (earnings) increases by age. After age 50 indirect costs declines because the
number of years to retirement age 65 becomes less than 15 years.
Projecting the costs onto the Swedish population
Sweden has a population of some 3.7 million persons in the ages 30-60, and we use the cost estimates presented
above to project onto this population and report the results in Table 7. The average annual hospital costs
accumulated over 15 years, is estimated to 2.1 billion SEK and practically all costs (97 %) can be attributed to
obese patients. Table 7 also reveals a marked gender differences in costs. The average annual hospital costs for
male patients are estimated to 1.36 billion (63 % of total) and 0.81 billion (37% of total) for female patients. In
Table 7 we also report our estimate of indirect costs due to death before retirement age. The indirect costs due to
overweigh and obesity have the same pattern as for hospital costs although a little higher. The average annual
indirect costs is estimated to 2.9 billion which is 38 % higher than our estimate of the annual costs of hospital
treatment for obese and overweight patients.
Discussion
In this study we have sought to provide answers to two questions. First we raised the question if hospital
treatment costs for obese and overweight are higher than for normal weight subjects. To this question we can say
yes. In our statistical analysis, we found the mean expected number of days in hospitals to be significantly higher
for obese and overweight patients than for normal weight patients, cf. Tables 5. This difference in number of
days of care in hospital we have calculated in Table 7 translates into an aggregate costs for the Swedish Health
Care Budget of some 2.1 billion SEK per annum. This figure we may now compare to another recently produced
study on the problems and costs of obesity in Sweden. SBU (The Swedish Council on Technology Assessment
of Health Care, 2003) have reviewed the literature on the problem and costs of obesity treatment and from this
10
review they assume that the costs of obesity is equal to 2 % of the total health care costs in Sweden [27]. This
assumption, which is based on literature review of costs estimates of overweight and obesity produced in other
countries, yields total costs of 3 billion SEK for the year 2002. This is higher than our cost estimate in this study
of 2.1 billion. The reason for this is that we in our study we have only estimated the total costs of all hospital
treatment and not included the costs of out-patients treatment including costs of drugs. In the Swedish health
care system around 60 % of the direct health care costs are for hospital care, which would imply that out of the 3
billion estimated by SBU 1.8 billion should be for hospital care and this is lower than our costs estimate of 2.1
billion. One explanation for this discrepancy between our estimate and the estimate provided by SBU is that we
in our study use panel date drawn from the population of Malmö city. Malmö is Sweden's third largest city and
access and utilization of health care (hospital care) may not be fully representative for the country population.
The discrepancy may also of course be due to differences in methodology. Our approach is a much more
rigorous approach that the simple approach used by SBU. Nevertheless, our approach and the approach taken by
SBU arrive at total costs estimates that are within reasonable margin comparability.
The second question we raised in this study was if there were any differences in survival for subjects with
different BMI. To this question we can also say yes. Obese subjects were found to have a significantly lower
probability of survival than normal weight subjects, cf. result Table 2. This difference in survival we have
calculated in Table 7 translate into an aggregate indirect cost of lost production due to early death for the
Swedish society of some 2.9 billion SEK per annum. For this estimate we have made the same assumption about
the prevalence as reported for direct costs. This is a weak assumption and therefore our result must be interpreted
with great care. Nevertheless, we believe the result gives a rough estimate of the magnitude of the indirect costs
involved for excess morbidity found for overweight and obese patients. However, our estimate is only part of
total indirect costs. There are three major reasons that give rise to indirect costs: (1) indirect costs due to death
before retirement, (2) indirect costs due to early retirement on account of morbidity and (3) indirect costs due to
short term absence from work on account of an illness episode. We have only estimated the indirect costs due to
death before retirement age. Unfortunately, our panel data set contains no information on early retirement on
account of morbidity and the extent of absence of work on account of short-term illness episodes. In Sweden it is
these two causes that give rise to most of the indirect costs. According to a cost of illness (COI) report published
by the National Board of Health and Social Welfare in Sweden, Socialstyrelsen (1996), cost of early retirement
and costs of short-term absence from work accounted for some 80 % of total indirect costs in Sweden [28]. If we
11
use this ratio on our estimated indirect cost of 2.9 billion SEK on account of death before retirement, we arrive at
a total indirect costs for overweight and obesity of some 14.5 billion SEK {= 2.9/(100%-80%)}.
The result in Table 5 shows the incremental costs per patient of obese and overweight subjects as compared to
normal weight. This incremental cost constitutes the potential costs savings that could be gained by any therapy
that could convert obese and overweight to normal weight subjects. However against this cost saving we must
deduct the cost of the therapy, which is likely to be equally as high if not higher, so the potential for any net
hospital costs saving to be gained seems to be rather limited. The indirect costs are substantially higher than
hospital costs so there seems to be a good potential to make savings in indirect costs with a therapy that could
convert obese and overweight subjects to normal weight subjects. However, to avoid the excess health-care cost
of obesity a primary strategy of prevention in children and young adults based on lifestyle is the most important
measure to undertake. Besides that, new drug treatment modalities should be developed for cost-effective
treatment of adult obesity with a favourable long-term effectiveness and safety profile.
A major limitation of our study is that we have not been able to control for differences in diseases among the
screened population. Nor did we in this study exclude subjects who reported that they smoked cigarettes or had a
history of cancer. Since cancer is associated with both low and high BMI and high medical costs, it might
therefore be a confounding factor to our estimation of costs. Smoking is primarily associated with low BMI and
is therefore another important confounding factor in our analysis. Excluding individuals with other confounding
factors, for example subjects with pre-existing coronary heart disease, stroke and type 2 diabetes, would have
resulted in a reduction of the study sub-population characterized by metabolic effects of obesity.
Another limitation of our study is that we have been unable to control for differences in socio-economic status
(SES) of the screened population. Unfortunately, the data set that we have used in this study contains insufficient
data to lend itself to analyze the link between SES and obesity. One study on the link between SES and obesity
concludes that the causality can run in both directions from low SES to obesity or from obesity to SES [30]. The
conclusion made in this study was therefore that the question on causality was too complex to lend it self to the
conclusion that obesity was a contributing factor for low SES. There are other common variables that link
obesity and low SES.
12
Most studies investigating the relationship between health care costs and BMI are based on the prevalence
approach and provide only a point estimate of excess costs for one single year. Such studies are, however, of
limited value for assessing the value of intervention programs. For evaluation of intervention programs cohort
studies, following patients over long periods and estimating the present value of potential cost savings, are more
relevant. To our knowledge, our study is the first to show a statistical significant relationship between BMI and
longitudinal inpatient care costs. Other cohort studies have established relationship between drug therapy costs
and outpatient care costs but failed in this respect for inpatient care costs [17,19]. However, the previous study
was based on a data set of 1286 subjects, selected from 7021 adults participating in a health survey, followed
over a period of 9 years. Our estimates were derived from a larger data set, in which 33332 subjects, selected
from a screening project involving 33346 individuals, were followed over 15 years.
In conclusion, we have found that obesity both entails a considerable health hazard and consequently gives rise
to substantial hospital costs and indirect costs due to early death. In particular we have found that inpatient care
costs are higher for overweight and obese patients indicating potential cost offsets especially for indirect costs by
an effective and safe weight loss intervention.
13
References
1. Pi-Sunyer FX. Medical hazards of obesity. Ann Intern Med 1993;119:655-60.
2. Calle EE, Thun MJ, Petrelli JM, Rodriguez C, Heath CW. Body-mass index and mortality in a prospective
cohort of U.S. adults. N Engl J Med 1999; 341:1097-105.
3. Wannamethee G, Shaper AG. Weight change in middle-aged British men: implications for health. Eur J Clin
Nutr 1990; 44:133-42.
4. Lee IM, Paffenberger RS Jr. Change in body weight and longevity. JAMA 1992; 268:2045-9.
5. WHO/NUT/NCD/98. Obesity. Preventing and managing the global epidemic. Report of WHO Consultation
on Obesity, Geneva, 3-5 June 1997.
6. Walker M, Wannamethee G, Whincup PH, Sharper AG. Weight change and risk of heart attack in middleaged British men. Int J Epidemiol 1995; 24:694-703.
7. Lean ME, Powrie JK, Anderson AS, Garthwaite PH. Obesity, weight loss, and prognosis in type 2 diabetes.
Diabet Med 1990;7:228-33.
8. Karason K, Wikstrand J, Sjöström L, Wendelhag I. Weight loss and progression of early atherosclerosis in the
carotid artery: a four-year controlled study of obese subjects. Int J Obesity 1999; 23:948-56.
9. Diabetes Prevention Program Research Group. Reduction in the incidence of type 2 diabetes with lifestyle
intervention or metformin. N Engl J Med 2002; 346:393-403.
10. Colditz GA. Economic costs of obesity. Am J Clin Nutr 1992; 55:503S-507S.
11. Levy E, Levy P, Le Pen C, Basdevant A. The economic cost of obesity: the French situation. Int J Obes Relat
Metab Disord 1995; 19:788-92.
12. Hughes D, McGuire A. A review of the economic analysis of obesity. Br Med Bull 1997; 53:253-63.
13. Kortt MA, Langley PC, Cox ER. A review of cost-of-illness studies on obesity. Clin Ther 1998; 20:772-9.
14. Kurscheid T, Lauterbach K. The cost implications of obesity for health care and society. Int J Obes Relat
Metab Disord 1998; 22 (suppl.1):S3-S5.
15. Wolf AM, Colditz GA. Current estimates of the economic cost of obesity in the United States. Obes Res
1998; 6:97-106.
16. Birmingham CL, Muller JL, Palepu A, Spinelli JJ, Anis AH. The cost of obesity in Canada. C MAJ 1999;
160:483-8.
17. Thompson D, Edelsberg J, Colditz GA, Bird AP, Oster G. Lifetime health and economic consequences of
obesity. Arch Intern Med 1999; 159:2177-83.
18. Thompson D, Brown JB, Nichols GA, Elmer PA, Oster G. Body mass index and future health care costs: A
retrospective cohort study. Obesity Research 2001; 9:210-8.
19. Oster G, Edelsberg J, O’Sullivan AK, Thompson D. The clinical and economic burden of obesity in a
managed care setting. Am J Manag Care 2000; 6:681-9.
20. Hartunian N S, Smart C I, Thompson M S. The Incidence and Economic Costs of Major Health Impairment.
Lexington Books. Toronto, 1981.
14
21. Berglund G, Nilsson P, Eriksson K-F, Nilsson J-Å, Hedblad B, Kristensson H, Lindgärde F. Long-term
outcome of the Malmö Preventive Project: Mortality and cardiovascular morbidity. J Internal Med 2000; 247:1929.
22. Norinder A, Persson U, Nilsson P, Nilsson J-Å, Hedblad B, Berglund B. Costs for screening, intervention
and hospital treatment generated by the Malmö Preventive Project; a large-scale community screening
programme. J Internal Med 2002; 251:44-52.
23. Jonsson S, Hedblad B, Engström G, Nilssson P, Berglund G, Janzon L. Influence of obesity on
cardiovascular risk. 23 years follow-up of 22 025 men from an urban Swedish population (accepted Int J Obesity
2002)
24. Nilsson P, Nilsson J-Å, Hedblad B, Berglund G, Lindgärde F. The enigma of increased mortality associated
with weight-loss in middle-aged men - the Malmö Prevention Project. J Internal Med 2002 (in press).
25. Duan, N. (1983) Smearing estimate: a nonparametric retransformation method. Journal of the American
Statistical Association, 78, 605-610.
26. R Development Core Team (2003). R: A language and environment for statistical computing. R Foundation
for Statistical Computing, Vienna, Austria. ISBN 3-900051-00-3, URL http://www.R-project.org.
27. Statistical Yearbook of Sweden 2002, Statistics Sweden.
28. SBU, The Swedish Council on Technology Assessment in Health Care.Fetma- problem och åtgärder. en
ststematisk litteraturöversikt. (Obesity- problem and interventions. A literature assessment), SBU, Stockholm
2002.
29. Socialstyrelsen, The National Board of Health and Welfare. Vad kostar sjukdomarna? (What is the costs of
illness?), Stockholm, 1996
30. Sörensen T.I.A. Socio-economic aspects of obesity: cause or effects? International Journal of Obesity, vol.
19. supplement 6, Nov. 1995.
15
Table 1 Baseline characteristics of subjects (N)
N
Age group
N<45
N 45-54
N >=55
Age range
min
max
Normal weight
Male
Female
12823
6662
52,7%
43,1%
4,2%
100,0%
28,3%
40,0%
31,7%
100,0%
Overweight
Male
Female
7931
2844
42,9%
50,5%
6,6%
100,0%
12,1%
44,4%
43,5%
100,0%
Obese
Male
Female
1382
1068
38,5%
54,7%
6,8%
100,0%
10,1%
40,2%
49,7%
100,0%
Age
BMI
25.5
28.2
27.9
28.2
28.1
28.5
61.2
57.6
61,1
57.4
60.8
57.0
Mean(SD) Mean(SD) Mean(SD) Mean(SD) Mean(SD) Mean(SD)
43.0(6.6) 48.6(7.9) 44.7(5.4) 51.7(5.5) 45.3(6.1) 52.2(5.4)
22.6(1.6) 22.1 (1.7) 26.9(1.3) 27.0(1.4) 32.4(2.4) 33.4(3.4)
Follow-up
years
18.6(3.6)
14.5(4.5
18.6(3.7)
13.2 (4.0) 17.9(4.5)
12.8(4.0)
16
Table 2 Parameter estimates for survival time, equation (2) estimated with N= 32710 patients
Value
Std. Error z value p value
Pr{>|z|}
α1 (Intercept)
αAge
αBMI = Normal
7.2398
-0.0571
0.10714
67.57 <0.0001
0.00175 -32.69 <0.0001
0
αBMI = Obese
-0.2472
0.03306
-7.48 <0.0001
αBMI = Overweight
-0.0154
0.02112
-0.73
αSex = Female
0.4610
0
αSex = Male
-0.5039
0.02990 -16.85 <0.0001
Log(scale)
-0.5286
0.01573 -33.61 <0.0001
Loglik (model) -20828.5
Chisq.
1567.07
Likelihood test ratio vs intercept Chisq. (χ2) = 1567, p value < 0,0001
Table 3 Parameter estimates probability of having non-zero hospitalization rate,
equation (3) estimated with N= 32710 patients (459511 patient-years)
Estimate
Std. Error z value p-value OR§)
Pr{>|z|}
β1 (Intercept)
-4.112957
βAge
0.0306225 0.0006842
βBMI = Normal
βBMI = Obese
0.03995 -103.36 <0.0001
44.76 <0.0001 1.031
0
0.4182596 0.0181195
23.08 <0.0001
1.52
βBMI = Overweight 0.1276276 0.0112628
11.33 <0.0001
1.14
14.00 <0.0001
1.18
βSex = Female
βSex = Male
0
0.1668386 0.0119167
Likelihood test ratio vs intercept Chisq. (χ2) = 2852, p value < 0,0001
§) OR (odds ratio) =expβ
17
Table 4 Parameter estimates for the log of # of days in hospital,
equation (4) estimated with N= 18153 patients (41228 patient-years)
Estimate
Std. Error
t value
p value
Pr {>|t|}
γ 1 (Intercept)
0.9222788
0.0456287
20.213
<0.0001
γAge
0.0147012
0.0007916
18.571
<0.0001
0.0659055
0.0216860
3.039
0.0024
γBMI = Overweight -0.0412887
0.0137294
-3.007
0.0026
0.0141413
0.738
0.46053
γBMI = Normal
γBMI = Obese
γSex = Female
γSex = Male
Adjust.R2
0
0
0.0104361
0.0091
18
Table 5: Mean (expected) days in hospital in normal weight, overweight and obese subjects, and excess days and
costs of excess days in hospital for overweight and obese subjects as compared to normal weight (mean and 95%
confidence intervals), by gender and selected age, over 15 years
Expected days in
hospital
Mean days per patient
Excess days in hospital
Mean days per patient (95% CI) §§)
Over-weight
Obese
Age§) Normal Over- Obese
weight
30
35
40
45
50
55
60
30
35
40
45
50
55
60
male
10.0
12.4
15.3
18.7
22.7
27.1
31.7
female
8.5
10.6
13.1
16.2
19.9
24.2
29.2
10.9
13.4
16.5
20.1
24.3
29.0
33.8
15.8
19.3
23.6
28.5
34.0
39.8
45.1
0.82
0.99
1.20
1.43
1.68
1.93
2.14
(0.27
(0.32
(0.37
(0.43
(0.47
(0.50
(0.48
1.35)
1.64)
1.98)
2.37)
2.80)
3.24)
3.65)
5.72
6.94
8.33
9.84
11.37
12.69
13.43
(4.10
(5.00
(6.02
(7.13
(8.23
(9.16
(9.61
7.62)
9.21)
11.02)
13.00)
15.02)
16.79)
17.88)
9.2
11.4
14.2
17.4
21.4
26.0
31.3
13.5
16.6
20.5
25.1
30.5
36.7
43.5
0.71
0.86
1.05
1.27
1.53
1.81
2.10
(0.22
(0.27
(0.32
(0.38
(0.44
(0.49
(0.54
1.17)
1.43)
1.75)
2.12)
2.55)
3.03)
3.55)
4.94
6.05
7.37
8.90
10.63
12.47
14.26
(3.47
(4.29
(5.27
(6.40
(7.66
(9.00
(10.28
6.65)
8.10)
9.82)
11.80)
14.05)
16.45)
18.82)
Cost*) of excess days in hospital
Mean cost per patient (95% CI)
Overweight
Obese
Age
male
30
35
40
45
50
55
60
2 369
2 883
3 479
4 157
4 897
5 650
6 328
(778 3 904)
(928 4 759)
(1 085 5 759)
(1 248 6 907)
(1 394 8 177)
(1 490 9 503)
(1 465 10 761)
16 567
20 130
24 220
28 733
33 378
37 572
40 350
(11 874
(14 507
(17 514
(20 813
(24 172
(27 142
(28 960
22 096)
26 731)
32 052)
37 946)
44 051)
49 653)
53 576)
female
30
2 042
(643 3 386) 14 295 (10 032 19 276)
35
2 502
(779 4 151) 17 527 (12 416 23 486)
40
3 051
(936 5 068) 21 371 (15 259 28 479)
45
3 693 (1 100 6 146) 25 857 (18 570 34 296)
50
4 434 (1 277 7 401) 30 945 (22 312 40 902)
55
5 265 (1 451 8 825) 36 452 (26 329 48 082)
60
6 149
(1 588 10 367) 41 941 (30 257 55 310)
§) We selected the age span 30 - 60 years because this age span covers the 95 % CI interval of age in our sample
population.
§§) We use Boots-trap techniques to generate our 95 % CI
*) discounted at 3 %
19
Table 6: Mean (expected) lost years of production in normal weight, overweight and obese subjects, and excess
years and costs of excess years of lost production for overweight and obese subjects as compared to normal
weight (mean and 95% confidence intervals), by gender and selected age, over 15 years
Expected lost years of
production
Excess lost years of production
Mean lost years per patient (95% CI) §§)
Mean lost years per
patient
Age§) Normal Over- Obese
weight
30
35
40
45
50
55
60
30
35
40
45
50
55
60
Male
0.10
0.16
0.26
0.42
0.66
0.35
0.08
Female
0.04
0.07
0.11
0.18
0.29
0.15
0.03
Overweight
Obese
0.10
0.16
0.26
0.43
0.68
0.36
0.08
0.15
0.24
0.39
0.62
0.99
0.52
0.11
0.00
0.00
0.01
0.01
0.02
0.01
0.00
(0.00
(-0.01
(-0.01
(-0.02
(-0.03
(-0.02
(0.00
0.01)
0.02)
0.03)
0.04)
0.07)
0.03)
0.01)
0.05
0.08
0.13
0.21
0.33
0.17
0.04
(0.04
(0.06
(0.10
(0.15
(0.24
(0.12
(0.03
0.06)
0.10)
0.16)
0.26)
0.41)
0.22)
0.05)
0.04
0.07
0.11
0.18
0.30
0.15
0.03
0.06
0.10
0.17
0.27
0.44
0.23
0.05
0.00
0.00
0.00
0.00
0.01
0.00
0.00
(0.00
(0.00
(-0.01
(-0.01
(-0.01
(-0.01
(0.00
0.00)
0.01)
0.01)
0.02)
0.03)
0.02)
0.00)
0.02
0.04
0.06
0.09
0.15
0.08
0.02
(0.01
(0.02
(0.04
(0.06
(0.10
(0.05
(0.01
0.03)
0.05)
0.07)
0.12)
0.19)
0.10)
0.02)
Cost*) of additional lost years of production
Mean cost per patient (95% CI)
Overweight
Obese
Age
Male
30
35
40
45
50
55
60
Female
30
35
40
45
50
55
60
Note: same as Table 5
844
1 251
2 256
3 609
5 281
2 957
748
(-1 140 3 186) 15 212
(-2 304 4 992) 25 765
(-3 594 8 727) 43 559
(-5 933 14 107) 69 751
(-8 948 20 921) 102 662
(-5 320 11 887) 58 854
(-1 275 2 998) 14 436
(10 712
(18 167
(31 431
(50 524
(74 536
(42 392
(10 242
19 155)
32 128)
54 253)
86 766)
127 903)
73 486)
18 143)
331
450
615
1 228
1 848
1 033
247
(-266
(-716
(-1 532
(-1 879
(-2 686
(-1 547
(-375
(3 230
(5 763
(9 619
(15 846
(23 515
(13 255
(3 141
6 867)
11 737)
19 377)
30 910)
45 077)
25 245)
6 007)
1 157)
1 720)
2 623)
4 701)
7 018)
3 948)
934)
5 129
8 882
14 814
23 640
34 552
19 435
4 629
20
Table 7: Average annual projected hospital and indirect cost (projected over 15 years), SEK mil, of overweight
and obese male and female persons in the Swedish population in ages 30-60 as compared to the same population
with normal weight.
Overweight
Hospital cost Indirect costs
(HC)
(PL)
Total
male
39,526
26,663
66,190
female
40,215
9,193
49,407
Total
66,190
35,856
102,045
Obese
HC
LP
Total
male
1 320,654
2 387,908
3 708,562
female
768,904
511,690
1 280,594
Total
2 089,558
2 899,598
4 989,156
Overweight + Obese
HC
LP
Total
male
1 360,181
2 414,571
3 774,752
female
809,118
520,883
1 330,001
Total
2 155,747
2 935,454
5 091,202
Note: Indirect cost measures loss of production due to early death. Costs are discounted using a 3 % discount
rate
21
Figure 1: Estimated survival in male subjects, aged 50
(Note: the vertical scale is limited to the range 0.7 to 1.0).
22
Figure 2: Estimated probability of hospitalization in male subjects, aged 50
23
Figure 3: Predicted number of days in hospital in male subjects, aged 50
24
Figure 4: Expected number of accumulated days in hospital by age at beginning of period
25