Risk Adjustment Algorithms

Risk Adjustment Algorithms

medicaid-utils includes Python implementations of 8 published clinical algorithms for risk adjustment, procedure classification, and quality measurement.

Elixhauser Comorbidity Index

Flags 31 comorbidity groups from diagnosis codes (Elixhauser et al., 1998). The implementation extends the original 30 categories by splitting hypertension into uncomplicated/complicated.

from medicaid_utils.adapted_algorithms.py_elixhauser.elixhauser_comorbidity import score

# MAX — first construct LST_DIAG_CD from individual diagnosis columns
diag_cols = [c for c in ip.df.columns if c.startswith("DIAG_CD_")]
ip.df = ip.df.map_partitions(
    lambda pdf: pdf.assign(
        LST_DIAG_CD=pdf[diag_cols].apply(
            lambda row: ",".join(v for v in row if v and str(v).strip()), axis=1
        )
    )
)
df_scored = score(ip.df, lst_diag_col_name="LST_DIAG_CD", cms_format="MAX")

# TAF — gather diagnosis codes (creates LST_DIAG_CD on dct_files["base_diag_codes"])
# ip.gather_bene_level_diag_ndc_codes()
# df_scored = score(ip.dct_files["base_diag_codes"], lst_diag_col_name="LST_DIAG_CD", cms_format="TAF")

Adds 31 binary columns (ELX_GRP_1 through ELX_GRP_31) to the DataFrame.

Reference: Elixhauser A, Steiner C, Harris DR, Coffey RM. Comorbidity measures for use with administrative data. Medical Care. 1998;36(1):8-27.

CDPS-Rx Risk Adjustment

Pharmacy-based risk adjustment using the Chronic Illness and Disability Payment System (Kronick et al., 2000, UC San Diego). Combines diagnosis-based CDPS categories with pharmacy (NDC) information.

from medicaid_utils.adapted_algorithms.py_cdpsmrx import cdps_rx_risk_adjustment

df_risk = cdps_rx_risk_adjustment.cdps_rx_risk_adjust(
    df, lst_diag_col_name="LST_DIAG_CD", lst_ndc_col_name="LST_NDC"
)

Reference: Kronick R, Gilmer T, Dreyfus T, Lee L. Improving health-based payment for Medicaid beneficiaries: CDPS. Health Care Financing Review. 2000;21(3):29-64.

BETOS Procedure Classification

Assigns Berenson-Eggers Type of Service categories to procedure codes using CMS crosswalk files.

from medicaid_utils.adapted_algorithms.py_betos import betos_proc_codes

df_classified = betos_proc_codes.assign_betos_cat(ot.df, year=2012)

Categories include evaluation & management, procedures, imaging, tests, DME, other, and exceptions.

ED Prevention Quality Indicators (ED PQI)

Flags potentially preventable emergency department visits (Davies et al., 2017).

from medicaid_utils.adapted_algorithms.py_ed_pqi.ed_pqi import get_ed_pqis

# Requires IP, OT, PS, and ED claim DataFrames
df_ed = ot.df[ot.df["ed_use"] == 1]
df_pqi = get_ed_pqis(df_ip=ip.df, df_ot=ot.df, df_ps=ps.df, df_ed=df_ed)

Reference: Davies S, Schultz E, et al. Identification of potentially preventable emergency department visits using a claims-based standardized methodology. Health Services Research. 2017;52(5):1667-1691.

Inpatient PQI (IP PQI)

AHRQ Prevention Quality Indicators for inpatient admissions. Identifies potentially avoidable hospitalizations that may indicate inadequate access to outpatient care.

from medicaid_utils.adapted_algorithms.py_ip_pqi.prevention_quality_indicators import pqirecode

df_adult, df_children = pqirecode(ip.df)

Returns separate DataFrames for adult (TAPQ01–TAPQ16) and pediatric (TAPD14–TAPD18) PQI indicators.

NYU/Billings ED Classification

Classifies ED visits by severity and preventability (Billings, Parikh, Mijanovich, 2000). Assigns probabilities to each visit across categories: non-emergent, emergent but primary-care treatable, emergent but preventable/avoidable, and emergent not preventable.

from medicaid_utils.adapted_algorithms.py_nyu_billings.billings_ed import get_nyu_ed_proba

pdf_nyu = get_nyu_ed_proba(
    df_ed, date_col="srvc_bgn_date", index_col="MSIS_ID", cms_format="MAX"
)

Reference: Billings J, Parikh N, Mijanovich T. Emergency department use: the New York story. Issue Brief (Commonwealth Fund). 2000;434:1-12.

Pediatric Medical Complexity Algorithm (PMCA)

Classifies pediatric patients by medical complexity into three tiers: non-chronic, non-complex chronic, and complex chronic (Simon et al., 2014, Seattle Children's Research Institute).

from medicaid_utils.adapted_algorithms.py_pmca.pmca import pmca_chronic_conditions

df_pmca = pmca_chronic_conditions(df, diag_cd_lst_col="LST_DIAG_CD_RAW")

Reference: Simon TD, Cawthon ML, Popalisky J, et al. Pediatric medical complexity algorithm: a new method to stratify children by medical complexity. Pediatrics. 2014;133(6):e1647-e1654.

Low-Value Care

Identifies low-value care services — clinical interventions with little or no benefit relative to their cost or potential harm (Charlesworth et al., 2016).

from medicaid_utils.adapted_algorithms.py_low_value_care.low_value_care import construct_low_value_care_measures

pdf_lvc = construct_low_value_care_measures(
    state="AL", year=2012,
    lst_bene_msis_filter=[], index_col="BENE_MSIS",
    max_data_root="/data/max", out_folder="/output"
)

Reference: Charlesworth CJ, Meath THA, Schwartz AL, McConnell KJ. Comparison of low-value care in Medicaid vs commercially insured populations. JAMA Internal Medicine. 2016;176(7):998-1004.

Algorithm Summary

Algorithm	Reference	Module
Elixhauser Comorbidity Index	Elixhauser et al., 1998	py_elixhauser
CDPS-Rx Risk Adjustment	Kronick et al., 2000	py_cdpsmrx
BETOS Classification	CMS	py_betos
ED PQI	Davies et al., 2017	py_ed_pqi
IP PQI	AHRQ	py_ip_pqi
NYU/Billings ED Algorithm	Billings et al., 2000	py_nyu_billings
PMCA	Simon et al., 2014	py_pmca
Low-Value Care	Charlesworth et al., 2016	py_low_value_care