Claims Simona Dalle Carbonare a, b , Fulvia Folli c , Emanuele - - PowerPoint PPT Presentation

Laboratory for Biomedical Informatics

UNIVERSITY OF PAVIA

A Methodology for the Extraction of Quantitative Risk Indices from Medical Injuries Compensation Claims

Simona Dalle Carbonarea, b, Fulvia Follic, Emanuele Patrinic,d, Riccardo Bellazzia

a Dipartimento di Informatica e Sistemistica, University of Pavia Italy b Institute for Advanced Studies, IUSS, Pavia, Italy c A. O. Lodi, Italy d Healthcare risk management, MARSH Italia

Agenda

! Topic introduction ! Actuarial Model ! Methodology description

! Variables ! Steps ! Extracted risk indices

! Application to real data ! Results ! Conclusions

Properly negotiate the contract with insurance companies Plan actions to mitigate risks Learn the main reasons causing the adverse events happening Analyze the medical injuries compensation claims

Introduction

One of the main goals of risk management in health care is to lower the frequency and the impact of the adverse events

Risk Management activities Revise insurance policy

Methodology Proposed

Methodology based on the actuarial model to obtain quantitative risk indices from compensation claims

Stratified analysis All data (claims) Global analysis

Number

• f event
• ccurrences

Impact of the event

Actuarial Model

Convolution Monte Carlo Simulation Estimated Losses L Frequency F Severity S

Model Variables

Number of claims in a given time interval Reimbursement of the claim (€) Severity Continuous Frequency Discrete

Methodology Steps

Estimation of frequency and severity distributions 1) Monte Carlo simulation to estimate the losses distribution 2) Risk indices extraction 3)

1) Distributions Estimation

Parametric fitting

Poisson Negative bin. Lognormal Exponential Frequency Discrete Severity Continuous

Statistical test

χ2 Kolmogorov Smirnov

Use the estimated parametric distribution resulting in the best adaptation

Good fitting

pvalue>0.05

Bad fitting

pvalue<0.05

Non-parametric estimation

Roulette Wheel Kernel Density Estimation

Bayesian Hierarchical model

2) Severity extraction

2) Monte Carlo Simulation

1) Frequency extraction 3) Sum of severity values extracted Single scenario {1,500} € {2} {1,000 500} € Estimated Losses Distribution L A lot of simulations…

A lot of simulations…

3) Risk Indices Extracted

Estimated losses distribution

Var - Expected Value Unexpected Losses 99th percentile Maximum expected loss Value at Risk (Var) Expected Value Mean Median

The data

Hospitals in the area

• f Lodi (Milan, Italy)

From 1999 to the first semester of 2007 Time period covered by data Data Number of compensation claims 317 claims

excluded excluded 110 still under evaluation 1 without compensation 206 with compensation included

Global Analysis Results

! Var  maximum expected losses ! Useful for insurance policies

The HCO deals with expected losses Insure only unexpected losses Reduction of the insurance premium Immediately benefit from the results of risk reduction strategies

Hospital supplying the health performance Hospital division

Stratified Analysis – Attributes

Location Risk Area Two variables of interest

To infer the shape of the losses distribution To better understand the nature of the adverse events

• f a hospital or division

To compare the risk indices with the amount

• f services supplied

To perform a fair comparison between hospitals and divisions

Stratified Analysis – Indices

Supplementary indices

Poor management

• f the exceptional events

Few events with large compensation high Presence of systematic errors Events related on average with high compensation

Var/Losses Median Indicator

low

Obstetrics and Gynecology Surgery

Risk Area Results

Risk Area Yearly Var € Losses Mean € Losses Median € Var/Losses Median UL/number

• f beds

Obstetrics and Gynecology 1,691,000 149,100 42,658 40 22,027 Surgery 1,129,700 202,570 134,030 8 3,622 Medicine 689,010 74,840 32,353 21 1,310 Services 420,690 34,773 6,378 66 N/A First Aid 379,470 62,160 37,941 10 N/A Others 81,471 14,047 8,327 10 503

Location Results

Location Yearly Var € Losses Mean € Losses Median € Var/Losses Median UL/number

• f beds

Hospital 1 1,412,000 261,210 178,240 8 2,615 Hospital 3 1,334,300 157,770 52,800 25 12,517 Hospital 2 122,860 35,143 29,746 4 467 Hospital 4 441,830 54,282 25,157 18 1,872 Others 91,766 9,669 2,610 35 N/A

Conclusions

! We proposed a probabilistic approach based on the

actuarial model to perform a risk analysis on data coming from medical injuries compensation claims

! The methodology has been applied to the data of

the healthcare organization of Lodi and allowed to:

! analyze the compensation claims and suggest some

insurance policy

! analyze the claims of single hospitals and divisions

identifying those associated with the major risk

! This is the first step towards the definition and

application of effective actions to prevent adverse events

Thank you for listening!

Laboratory for Biomedical Informatics

UNIVERSITY OF PAVIA

A Methodology for the Extraction of Quantitative Risk Indices from Medical Injuries Compensation Claims

Simona Dalle Carbonarea, b, Fulvia Follic, Emanuele Patrinic,d, Riccardo Bellazzia

a Dipartimento di Informatica e Sistemistica, University of Pavia Italy b Institute for Advanced Studies, IUSS, Pavia, Italy c A. O. Lodi, Italy d Healthcare risk management, MARSH Italia