arnar berþórsson ba kristlaug h. jónasdóttir bs, msc
DESCRIPTION
Recommendations on Minimum Data Recording Requirements in Hospitals from the Directorate of Health in Iceland: Is it possible to use Hospital Patient Registry data to decrease the cost of outliers. Arnar Berþórsson BA Kristlaug H. Jónasdóttir BS, MSc. Landspítali University Hospital (LSH). - PowerPoint PPT PresentationTRANSCRIPT
Recommendations on Minimum Data Recording Requirements in Hospitals from the Directorate of
Health in Iceland:Is it possible to use Hospital Patient Registry data to
decrease the cost of outliers
Arnar Berþórsson BAKristlaug H. Jónasdóttir BS, MSc
Landspítali University Hospital (LSH)Key statistics 2008
Population in Iceland 319.326
Number of individuals receiving hospital care 106.699
Outpatient units - visits 367.540Day units - visits 93.422Emergency department - visits 94.650Hospital at home service - visits 14.798
Admissions 28.607Patient days 232.570Average length of stay (LOS) 8,1
Average LOS Excluding Division of Rehabilitation and Division of Geriatrics 5,2Patient acuity 1,18
Deliveries 3.376Surgical procedures 14.583Diagnostic imaging 123.956
Number of employees (at the end of the year) 5.022Full-time equivalents (mean) 3.872
Outliers as persetages of total admissions 3.0%Outliers cost as a percentage of total operational cost 21,5%
Prospective Payment Systems (PPS) and Diagnosis Related Groups (DRG) Fixed payment per discharge. Payment is the same for all patients within each DRG
group. Patients within each DRG group should show homogeneity
in clinical conditions as well as in cost. Payment for DRG groups is based on average costs for
patient within the group. Patients grouped based on:
Principle diagnosis ICD-10 Secondary diagnosis ICD-10 Procedures and imaging examination NCSP+ Length of stay Age Gender Type of discharge
DRG weight: mean cost in each DRG divided by total mean cost in all DRGs.
Outliers An observation that is numerically distant from
the rest of the data. In most large samples of data, some data points will
be further away from the sample mean than what is deemed reasonable
They can occur by chance, but they can also be an indicator of either measurement- or coding errors or that the data has a heavy-tailed distribution.
In health care reimbursement, especially in PPS, outliers are those patients that require an unusually long hospital stay or whose stay generates unusually high costs.
Hypothesis
p measures the probability that a patient will become an outlier.
T0 :Following model, based on Guidelines from the Directorate of Health for minimal registration requirements for patient information, can be used as an indicator for a patient’s probability of becoming an outlier.
Log (p/ 1-p) =c+β1*gender+ β2*age (+70) +β3*age (0-18)+β4 * ln(Number of IDC-10 diagnosis)+β5* ln( Number of NCPS+ theraphutic procedures)+β6* Types of admissions+β7* Types of discharges_MORS+β8*Types of Discharges_Other+β9*Ln(LOS)+e
Calculation of outliers
Outliers are admissions that exceed a certain cost limits calculated within each DRG group, see formula below.
Outlieri = Q3i + k *(Q3i – Q1i)
k = (P95 – Q3) / (Q3 – Q1)
Where Q1 is 25th percentile, Q3 is 75th percentile and k is a constant that set the outlier limit to 5 percent. P95 is 95th percentile.
Methodology Research design: Non-experimental analytic
analysis. Sample: Discharges from all wards within LSH
except: Long term Geriatric wards Long term Psychiatric wards Rehabilitation wards Palliative care ward Healthy newborns
Sample criteria: Discharges in the period 1. Jan – 31. Des 2008 (n=21.912) Cases classified into DRG groups DRG groups ≥ 30 cases (196 DRG groups)
Data analysis: Logistic regression (stepwise method)
Methodology
Dependent variable: Outlier=1, Non Outlier=0 Independent variables :
Gender, 1=male, 2=female Age, children ≤ 18, adults 19 to 69, elderly ≥ 70 Number of ICD-10, (International Classification of
Deceases) codes, (Transformed to ln(x) to correct skewness)
Number of NCSP+ codes, (Nordic Classification of Surgical Procedures), (Transformed to ln(x) to correct skewness)
Types of admissions, acute =1, non acute =0 Types of discharges, home=1, died=2, other=3 Length of stay, (LOS) (Transformed to ln(x) to
correct skewness)
Methodology: Sample
Number percent Number of outliersTotal number 21.912 / 703 3%Gender Male 9.194 42% 322 3,5%
Female 12.718 58% 381 3,0%
Types of adamission Acute 17.494 80% 598 3,4%Non acute 4.418 20% 105 2,4%
Discharge Home 19.895 91% 406 2%Mors 341 2% 49 14%Other* 1.676 8% 248 15%*Nursing-homes, other hospitals and other institutes
Sample
Methodology
Logistic regression
predict the probability of Y occorrung given known values of predicting variables
Result
Acute admission* 1,77 0,01Length of stay* 1,94 0,01Number of ICD-10* -0,36 0,01Number of NCSP* 1,20 0,01Mors* 1,32 0,01Transferd* 0,46 0,0117 years and yonger* 0,78 0,0170 years and older** -0,30 0,001Constant -8,41 0,01
Change in risk
p<
Discussions Why is it that with increasing number of
registered diagnosis the probability of a patient becoming an outlier decreases??
Children (0-17) are more likely to become outliers than 18-69 years old
But older patients (70+) are less likely to become a outlier than 18-69 years old.
Death, mortality and length of stay provide strong evidence of who become an outliers.
Patient that are discharged to nursing homes, other hospitals and institutes are more likely to become an outlier.
Limitation
DRG groups with fewer than 30 discharges were ignored.
Cost is partly distributed by Length of stay, does this cause problem for the assumption to the model?
We could not use Marital Status Distinguish between Discharges to other specialitis
and to other institutions.
Use of the result
The purpose is not to decrease outliersThe purpose is to influence the factors that cause
the patient to be a outlier.According to this study, outliers are 7 times more
expensive than average patient in the same DRG group.
Further studies and ideas
Effect of marital status and discharge mode Connection between number of registered diagnosis
and outliers within DRG group Add other relevant variables to the model such as
Acuity, re-admission, waiting list, chronic diseases, test results….
Limit the sample to smaller groups such as single DRG groups or MDC groups or speciality.
Effect of quality of coding and homogeneity of DRG groups.
Result I