Analysis of New York State Medicaid Program Enrollment by Month: Beginning 2009

Transcription

1 Analysis of New York State Medicaid Program Enrollment by Month: Beginning 2009 Stephanie Feddock, Jennifer Flynn, Michele Kirchhoff, Nader Nassar, James Sicuranza Seidenberg School of CSIS, Pace University, White Plains, New York Abstract evaluate the number of New York State Medicaid enrollees by eligibility year and month within each economic region, health insurance plan information, and demographics using data mining algorithms and tools. I. INTRODUCTION The Medicaid enrollment and eligibility data is a summary of Medicaid beneficiaries by calendar month and year. Included in the dataset are the geographic location, the aid category group, dual Medicaid/Medicare status, managed care status, health plan name if enrolled in managed care, the health plan type if enrolled in managed care, gender, age group and race/ethnicity [3]. Research questions include: Are there differences in the number of enrollees by race, age, gender, or economic region within NYS? What is the difference for dual eligibility by age? Most enrollments? Least enrollments? II. DATASET AND METHODOLOGY This dataset aggregates and displays the number of New York State Medicaid enrollees by eligibility year and month within each NYS Economic Region; health insurance plan information; and enrollee demographics [1,2]. The data depicts the number of Medicaid enrollees by year, month, economic region, aid category, dual eligibility status (Medicare and Medicaid), plan name, plan type, gender, age group, and race/ethnicity. The data contains twelve variables: eligibility year, eligibility month, economic region, aid category, dual eligible, Managed Care vs. Fee For Service, plan name, plan type, gender, age group, race/ethnicity, and the number of enrollees [3]. For this analysis a few scenarios were performed. During preprocessing the string data was converted to Nominal to run algorithms for analysis. To get a baseline for analysis the ZeroR was performed to classify the data and then the J48 algorithm was performed. SimpleKMeans was used to Cluster data and the Apriori algorithm was used to find trends. III. TOOLS For this analysis we used the Weka algorithms for the data mining tasks directly from the tool to the dataset. The tools used for this evaluation include preprocessing, classification, clustering, association rules, and visualization. For guidance Chapter 17 in the Data Mining Text book was used to perform the scenarios on the data. All results are attached in zip with text file results. Screen shots are included in this document. IV. ANALYSIS During preprocessing nine of the twelve attributes were converted to StringtoNominal for processing. The attributes include Eligibility Year, Eligibility Month, Economic Region, Aid Category, Dual Eligible, Managed Care vs. Fee For Service, Plan Name, Plan Type, Gender, Age Group, Race, and Number of Recipients. Eligibility Year, Eligibility Month, and Number of Recipients are numeric attributes, the rest were Strings converted to Nominal. There is a total of 619,437 instances in this dataset from Eligibility Years 2009 until The Number of Recipients represents the number of enrollees in the dataset. There were 12,087 distinct enrollees. All WEKA results have been converted to text and attached in the zip file. A5.1

2 Using the Classify Tab the ZeroR was run on Economy_Region, Plan_Tye, and Plan_Name to get a baseline. ZeroR predicted New York City for Economy_Region, HMO/PHSP for Plan_Type, and Fee-For-Service for Plan_Name. The J48 algorithm on the dataset shows that in the eleven distinct Economic_Regions the following results show most recipients are WHITE between the Age between The SimpleKMeans algorithm performed on the dataset was clustered into two Economic_Regions, New York City and Long Island. Showing results of White Females between the Age of use HMO/PHSP Plan Types with the most in New York City using the FEE-FOR-SERVICE Plan, followed by FIDELIS HC NY in Long Island. When eliminating RACE and GENDER and adding AID_CATEGORY and MCvFFS the results show New York City as the highest in Economic_Region using Managed Care, with SSI AID_CATEGORY, that are not Dual_Eligible on an HMO/PHSP Plan_Type using FEE-FOR-SERVICE Plan between the AGE of Using the Associate Tab the Aprioir Algorithm was performed on the dataset twice with nine then four attributes. The first result shows nine attributes with five large itemsets showing ten best rules. The best rule is Plan_Type=HMO/PHSP ==> ManagedCare_v_FFS=MMC A final set was run with only four attributes but no results were found. Visualization screen shots are below for the dataset. See additional details below. For this analysis the data was group by Plan_Type. There are ten distinct plans. Table1, below TABLE1 Data Analysis Plan_Type Count HMO/PHSP PARTIAL MLTC FEE-FOR-SERVICE PACE 9993 MEDICAID ADVANTAGE MAP 7715 SENIOR HMO 125 AIDS SNP PARTIAL CAP PLAN 9179 FFS PARTIAL CAP 1249 It was then grouped by Economic_Region. There are eleven distinct regions. TABLE2 Data grouped by Economic Region Economic_Region Count Mid-Hudson Mohawk Valley New York City North Country Other Southern Tier Western Capital District The second set was run with 6 attributes generating two large itemsets. The best rule is Plan_Type=FEE- FOR-SERVICE ==> Plan_Name=FEE-FOR- SERVICE Central Finger Lakes Long Island A5.2

3 The last was Plan_Name by Economic_Regions. Plan_Name has 84 distinct plans. This was then analyzed by Plan_Type to Economic_Region V. WEKA EVALUATION OF DATASET A Load Dataset Medicaid_Program_Enrollment_by_Month Begin ning_2009.arff file to Weka B- Preprocess Apply Filter: Unsupervised Attribute StringToNomial Instances = 619,437 Attributes = 12 o Eligibility _Year: Distinct 5: Numeric o Eligibility_Month : Distinct 12 : Nominal o Economic_Region: Distinct 11 : Nominal (See Table 1) Mid-Hudson Mohawk Valey New York City North Country Other Southern Tier A5.3

4 Western Capital District Central o Aid_Category: Distinct 3 : Nominal SSI TANF/SN FHPLUS o Dual_Eligible: Distinct 2 : Nominal Yes No o ManagedCare_v_FFS: Distinct 2 : Nominal MMC FFS o Plan_Name: Distinct 84 : v o Plan_Type: Distinct 10 : Nominal HMO/PHSP PARTIAL MLTC FEE-FOR-SERVICE PACE MEDICAID ADVANTAGE MAP SENIOR HMO AIDS SNP PATRIAL CAP PLAN Finger Lakes Long Island FFS PARTIAL CAP o Gender: Distinct 2 : Nominal Male Female o Age_Group: Distinct 4: Nominal o Race: Distinct 6 : Nominal WHITE ASIAN BLACK HISPANIC NATIVE AMERICAN OTHER o Number_of_Recipents: Distinct 12087: Numeric Relation: Medicaid Program Enrollment by Montho weka.filters.unsupervised.attribute. StringToNomial R 3-11 Visual - All Attributes C. Classify ZeroR predicted New York City for Economy_Region, HMO/PHSP for Plan_Type, and Fee-For-Service for Plan_Name. The J48 algorithm on the dataset shows that in the eleven distinct Economic_Regions the following results show most recipients are WHITE between the age of A5.4

5 VI. Economic_Region Race Age Mid-Hudson BLACK Mohawk Valley WHITE New York City HISPANIC North County WHITE OTHER WHITE Southern Tier BLACK Western WHITE Capital District WHITE Central WHITE Finger Lakes WHITE Long Island WHITE SUMMARY The evaluation to determine New York State Medicaid enrollees by health insurance plan information, economic region, and demographics produced a few interesting findings based on our analysis. Most New York State Medicaid enrollees from 2009 until 2013 were between the age of The highest amount of enrollees in given time was January 2013 in New York City using AID_CATEGORY TANF/SN with No DUAL using Managed Care. The Plan was HealthFirst Type HMO/PHSP, for Hispanic Females between There were 77,493 enrolled during that period. There were 105,507 instance of single enrollments by Month, Year, Region, Aid, Dual, MC_FFS, Name, Type, Gender, Age, Race and Enrollees The dataset can be analyzed by other attributes Some limitations is approximately a three months lag for eligibility information, meaning that the most complete and current data will always be at least three months old. This allows for the eligibility data to be complete at the time of the release. Medicaid enrollment, aid category assignment, and managed care enrollment are subject to change over time due to changes in beneficiary eligibility status [1]. Some future work questions: Are there differences in the number of enrollees by aid category, dual eligibility status, or Medicaid Managed Care vs. Fee For Service? How does the number of enrollees change over time? Does the proportion of enrollees by race, age, gender, or economic region change over time? Does the proportion of enrollees by aid category, dual eligibility status, or Medicaid Managed Care vs. Fee For Service change over time? Table4: New York State Counties that Make Up Economic Regions Economic Region Capital District Central Finger Lakes Long Island Mid-Hudson Mohawk Valley New York City North Country Southern Tier Western Other County Albany, Rensselaer, Saratoga, Schenectady, Warren, Washington, Columbia, Greene Oswego, Cayuga, Cortland, Madison, Onondaga Livingston, Ontario, Seneca, Wayne, Yates, Genesee, Monroe, Orleans, Wyoming Nassau, Suffolk Putnam, Rockland, Westchester, Dutchess, Orange, Sullivan, Ulster Fulton, Montgomery, Herkimer, Oneida, Otsego, Schoharie Bronx, Kings, New York, Queens, Richmond Clinton, Essex, Franklin, Hamilton, Jefferson, Lewis, St. Lawrence Chenango, Delaware, Tompkins, Broome, Chemung, Schuyler, Steuben, Tioga Allegany, Cattaraugus, Chautauqua, Erie, Niagara Unknown A5.5

6 VII DATA DICTIONARY The data set contains the following variables: Label Description Format & Length Eligibility Year Year, derived from varchar, length 4 the enrollee's date of Medicaid eligibility for respective year Eligibility Month Month, derived from varchar, length 2 the enrollee's date of Medicaid eligibility Economic Region NYS Economic varchar, length 16 region, based on the enrollee's county of residence Aid Category Medicaid aid varchar, length 7 category The aid categories are; Family Health Plus (FHPLUS), Social Security Supplemental Security Income (SSI), and Temporary Assistance for Needy Families / Safety Net (TANF/SN) Dual Eligible Flag indicating if enrollee is dually eligible for both varchar, length 3 Medicaid and Medicare Managed Care vs. Enrollment in varchar, length 3 Fee For Service Medicaid Managed Care or Fee For Service Plan Name Medicaid Managed varchar, length 29 Care Health Plan name Plan Type Medicaid Managed varchar, length 18 Care Health Plan. Health plan types are: Medicaid Advantage Plus (MAP), Medicaid Advantage, FFS Partial Capitation, Health Maintenance Organization/Prepai d Health Service Plan (HMO/PHSP), Partial Managed Long Term Care (Partial MLTC), AIDS Special Needs Plans (AIDS SNP), Partial Capitation Plan, and Senior HMO Gender Enrollee's gender varchar, length 6 Age Group Enrollee's age group varchar, length 5 Race Enrollee's race. varchar, length 15 Number of Recipients Number of Enrollees numeric, 5 Last Update March 19, 2014 REFERENCES [1] [2] Enrollment-by-Month-Beginning-200/m4hz-kzn3/about [3] 4AB2-822A-53796B04BD35?download=true [4] ence/mrg/glossary.pdf A5.6