Www.buseco.monash.edu.au/centres/che Centre for Health Economics Modelling the impact of being obese on hospital costs Katharina Hauck Bruce Hollingsworth.

www.buseco.monash.edu.au/centres/che

Centre for Health Economics

Modelling the impact of being obese on hospital costs

Katharina Hauck

Bruce Hollingsworth

A project funded by the NHMRC (grant number 334114) and the ARC (grant number DP0772235)

2



Background

• Cost of obesity (and related co-morbidities) to the health care

system are a concern

• Studies may underestimate the economic cost of obesity

• Obesity directly causes illnesses which are costly to treat

• Obesity may also influence the progression or severity of other

illnesses, including ones which are not directly caused by obesity

3



Research Question and Approach

• Is it more costly to treat obese patients, once they are in hospital?

• Difference in cost irrespective of type of illness and procedure?

• Analyse impact on length of stay (LOS) of inpatients

• LOS is major determinant of hospital costs

• Generate different estimates over the whole distribution of LOS

(from one night to very long)

4



Data

• Australian administrative public hospital data ‘Victorian Admitted

Episodes Data’ (VAED) for 2005/06

• Analysis on patient level

• Patient defined as obese if one of 2nd to 12th diagnosis code falls

within the range of ICD codes "E660“ to "E669“

• Our sample: financial year 2005/06 with 461,563 inpatients, of

which 6,086 (1%) are obese

5



Model

• LOS = f (obese, age, gender, nonelective, private payer, index of

social advantage, cost weight, number of diagnoses and

procedures, total separations of hospital, type and location of

hospital)

• Coefficient on dummy variable ‘obese’ is estimate of impact of

obesity (+ more costly, - less costly)

• Analysis for selected hospital specialties, and for medical and

surgical admissions

6



Problem: Outliers

• Problem: upper and lower outliers with respect to LOS

• In VAED: 3.4% of Patients stay very long and 1.3% very short,

conditional on observable characteristics

• Outlier status established with OLS regression of LOS on

explanatory factors

• Observations are

– Lower outliers if resOLS < Q(25) - 3*Inter Quartile Range

– Upper outliers if resOLS > Q(75) + 3*Inter Quartile Range

7



Estimation: Quantile Regression

• Problem: Large proportion of outliers violates assumptions of

normality of Ordinary Least Squares Regression

• Solution: Quantile regressions on 19 quantiles of LOS

• Quantiles of the conditional distribution of LOS are expressed

as functions of observed covariates

• Quantiles range from 0.05 (very short LOS) to 0.95 (very long

LOS), including the median 0.5

8



Estimation: Quantile Regression

• Quantile regression minimizes a sum of absolute residuals

• Residuals are weighed asymmetrically (for all quantiles except

the median)

– According to quantile, differing weights are given to positive and

negative residuals

• Outliers do not bias estimates at other quantiles

• Quantile regressions allow for differing impact of being ‘obese’ at

various points of the distribution of LOS

9



Summary statistics

Non-obese Obese

MeanStandard Deviation

MeanStandard Deviation

Length of stay 2.55 10.40 6.07 11.11

Cost weight 0.72 1.61 1.66 2.64

Age 53.61 23.38 58.58 15.36

Number of diagnoses 4.52 3.12 7.752 2.99

Number of procedures 2.40 2.49 3.01 2.77

Non-elective admissions 56.88 % 65.02 %

Medical admissions 68.45 % 66.15 %

Admissions to major teaching hospital 71.56 % 73.42 %

Admissions to city or big rural hospital 13.59 % 13.83 %

Privately paying patients 8.13 % 6.58 %

10



Results – Hospital Specialties

11



Results – Hospital Specialties

12



Results - Hospital Specialties

13




14




15




16




17




18



Results – Episode type

19



Results – Episode type

20



Why have obese different LOS?

• Why do obese stay longer in some specialties, but shorter in

others?

• Possible answers:

– Obese stay longer when they are treated as a medical case because

they are more complex?

– Obese stay shorter when they are treated as a surgical case because

they are much more complex, and are transferred to another hospital

(risk/cost shifting), or even die?

Any ideas?

21



Why have obese different LOS?

22



Future Research

• Investigate reasons for cost differences

• Analyse reasons for different patterns across specialties

• Use data on:

- Transfers to other hospitals

- Readmissions (to the same, and different hospitals)

- Complications and adverse events

- Mortality rates (in-hospital, and 30 day after stay)

23



Probit estimations

• Difference in probability of being transferred to another hospital

when obese, conditional on other explanatory factors

– Negative effect (?!) of ‘obese’ for Haematology, Respiratory and

Endocrinology, insignificant for all other specialties

• Difference in probability of dying when obese, conditional on other

explanatory factors

– Negative effect (?!) of ‘obese’ for the whole sample, and a range of

specialities including Orthopaedics, Cardiology,

General Medicine, and General Surgery.

Www.buseco.monash.edu.au/centres/che Centre for Health Economics Modelling the impact of being obese on hospital costs Katharina Hauck Bruce Hollingsworth.

Documents

los of inpatients los

health economics model

obese slide

obesity slide

health economics problem

long los

health economics estimation

short los