Handling silently missing data in Medicare Advantage encounter

  • Slides: 17
Download presentation
Handling silently missing data in Medicare Advantage encounter data Laura Hatfield, Ph. D Associate

Handling silently missing data in Medicare Advantage encounter data Laura Hatfield, Ph. D Associate Professor Department of Health Care Policy Harvard Medical School ICHPS - San Diego, CA January 7, 2020

MA plans care about diagnosis codes Risk-adjusted payments & bonuses Health information Fee-for-service payments

MA plans care about diagnosis codes Risk-adjusted payments & bonuses Health information Fee-for-service payments Diagnosis codes Procedure codes Quality data Diagnosis codes Procedure codes Premiums

Technological limitations • Provider capacity • Plan capacity Risk-adjusted payment incentive

Technological limitations • Provider capacity • Plan capacity Risk-adjusted payment incentive

Risk-adjustment is changing m e t s ym stm a t. P sk i

Risk-adjustment is changing m e t s ym stm a t. P sk i R o c En 2012 A en em ns o si is ) S D y t. S en u j Ad (R ) PS m b su (E st y a. S an r (T t a r. D e t n ti io pe d) o ri a o s st s ce rc a e es 2015 2016 2017 2018 S % 2019 t en 0 10 R 2014 S ac r he un 2013 ED t da 2020 ED k ris a tm s u dj

Encounter data quality may be poor Source: Med. Pac analysis, 2018

Encounter data quality may be poor Source: Med. Pac analysis, 2018

Distribution of the outcome, given other variables and the (random) missingness indicator Distribution of

Distribution of the outcome, given other variables and the (random) missingness indicator Distribution of the missingness indicator, given the other variables Distribution of the other variables Outcome variable of interest, which is sometimes missing Observation indicator variable, R=1 if outcome variable is observed, R=0 otherwise Other variables related to missingness indicator or outcome variable

Filling in the missingness indicators True Disease Diagnosis code Present True positive Present (R=1)

Filling in the missingness indicators True Disease Diagnosis code Present True positive Present (R=1) Absent False positive (R=1) (upcoding? ) False negative True negative (R=0) (R=? )

Auxiliary data sources

Auxiliary data sources

Thank you! @laura_tastic @HPDSLab www. Health. Policy. Data. Science. org hatfield@hcp. med. harvard. edu

Thank you! @laura_tastic @HPDSLab www. Health. Policy. Data. Science. org hatfield@hcp. med. harvard. edu