Presentation is loading. Please wait.

Presentation is loading. Please wait.

Analytical Example Using NHIS Data Files John R. Pleis.

Similar presentations


Presentation on theme: "Analytical Example Using NHIS Data Files John R. Pleis."— Presentation transcript:

1 Analytical Example Using NHIS Data Files John R. Pleis

2 Research Question Is the type of health insurance coverage held by adults > 65 years of age associated with flu shot use?

3 Additional Covariates Race/ethnicity Region of residence Education, marital status, sex Smoking Number of physician office visits Race/ethnicity Region of residence Education, marital status, sex Smoking Number of physician office visits

4 Additional Covariates Regular place of health care Selected chronic conditions diabetes, respiratory difficulties, or heart disease Low-income program participation Regular place of health care Selected chronic conditions diabetes, respiratory difficulties, or heart disease Low-income program participation

5 Data Files Determine which data files are needed for the analysis A good source for determining the file content is the Survey Description document: http://www.cdc.gov/nchs/nhis.htm http://www.cdc.gov/nchs/nhis.htm Determine which data files are needed for the analysis A good source for determining the file content is the Survey Description document: http://www.cdc.gov/nchs/nhis.htm http://www.cdc.gov/nchs/nhis.htm

6 Data Files This analysis will utilize data from several files, which include: Person Sample adult Family This analysis will utilize data from several files, which include: Person Sample adult Family

7 Person File Each person record also has a sampling weight Used to inflate each observation Adjusted for non-response as well as U.S. Census population totals by age, sex, and race/ethnicity Each person record also has a sampling weight Used to inflate each observation Adjusted for non-response as well as U.S. Census population totals by age, sex, and race/ethnicity

8 Person File Sum of the weights = Size of the Civilian Non-Institutionalized Population For more information regarding weights and other design issues, please attend: Practical Applications in Design and Analysis of Complex Sample Surveys (Session # 30) Sum of the weights = Size of the Civilian Non-Institutionalized Population For more information regarding weights and other design issues, please attend: Practical Applications in Design and Analysis of Complex Sample Surveys (Session # 30)

9 Sample Adult File Each sample adult record has a sampling weight Different from the person sampling weight Sum of the weights = Size of the Civilian Non-Institutionalized Population of adults > 18 years of age Each sample adult record has a sampling weight Different from the person sampling weight Sum of the weights = Size of the Civilian Non-Institutionalized Population of adults > 18 years of age

10 Sampling Weights Each data file has its own sampling weights Weights should be used, if not: Totals, means, and proportions are affected Estimates such as regression coefficients are biased Each data file has its own sampling weights Weights should be used, if not: Totals, means, and proportions are affected Estimates such as regression coefficients are biased

11 2000: Race/ethnicity (%) Sample Adults (aged > 65) Source: 2000 NHIS RACE/ ETHNICITY UNWEIGHTEDWEIGHTED Hispanic 9.3 5.9 NH White77.883.9 NH Black11.3 8.2

12 Sample Design The NHIS has a complex sample design The sample design affects the computation of variance of estimates A complex sample will produce larger variances than a Simple Random Sample (SRS) The NHIS has a complex sample design The sample design affects the computation of variance of estimates A complex sample will produce larger variances than a Simple Random Sample (SRS)

13 Sample Design Compared to a SRS, confidence intervals are wider, and statistical significance is harder to achieve for complex survey data If variance estimates are needed, the complex sample design should be accounted for in the analysis Compared to a SRS, confidence intervals are wider, and statistical significance is harder to achieve for complex survey data If variance estimates are needed, the complex sample design should be accounted for in the analysis

14 NHW vs. NHB Men Aged < 65: Bed Days (1)(2)(3) SRS Unweighted SRS Weighted Complex Weighted NHWNHBNHWNHBNHWNHB Mean3.425.193.525.363.525.36 S.E. of Mean.23.78.28.89.281.01 t-stat2.181.961.79 Sig. Level p =.0293p =.0497p =.0738 Source: 2000 NHIS

15 Research Question Is the type of health insurance coverage held by adults > 65 years of age associated with flu shot use?

16 Additional Covariates Race/ethnicity Region of residence Education, marital status, sex Smoking Number of physician office visits Race/ethnicity Region of residence Education, marital status, sex Smoking Number of physician office visits

17 Additional Covariates Regular place of health care One place that the adult usually went to when either sick care or preventive health care was needed Does not include emergency rooms (< 0.5% of the sample) Regular place of health care One place that the adult usually went to when either sick care or preventive health care was needed Does not include emergency rooms (< 0.5% of the sample)

18 Additional Covariates Respiratory difficulties Asthma (EVER) Chronic Obstructive Pulmonary Disease (COPD) Respiratory difficulties Asthma (EVER) Chronic Obstructive Pulmonary Disease (COPD)

19 Additional Covariates Heart disease (EVER) Coronary heart disease Angina pectoris Heart attack Any other heart condition Heart disease (EVER) Coronary heart disease Angina pectoris Heart attack Any other heart condition

20 Additional Covariates Low-income programs Supplemental Security Income Temporary Assistance for Needy Families (TANF) Food stamps Governmental rental assistance Low-income programs Supplemental Security Income Temporary Assistance for Needy Families (TANF) Food stamps Governmental rental assistance

21 Creating the File Not all the variables of interest for this analysis are contained in one file The Person, Sample Adult, and Family files can be merged to create one data file Not all the variables of interest for this analysis are contained in one file The Person, Sample Adult, and Family files can be merged to create one data file

22 Creating the File Person file Health insurance Race/ethnicity (all) Governmental rental assistance (last 12 months) Person file Health insurance Race/ethnicity (all) Governmental rental assistance (last 12 months)

23 Creating the File Sample Adult file Flu shot use (last 12 months) Race/ethnicity (partial) Smoking, chronic conditions Number of physician office visits (last 12 months) Sample Adult file Flu shot use (last 12 months) Race/ethnicity (partial) Smoking, chronic conditions Number of physician office visits (last 12 months)

24 Creating the File Sample Adult file Sample Adult weight Sample Adult file Sample Adult weight

25 Creating the File Family file Any family member received any of the following in the past 12 months: Supplemental Security Income TANF Food stamps Family file Any family member received any of the following in the past 12 months: Supplemental Security Income TANF Food stamps

26 Creating the File Person and Sample Adult files Education, marital status, sex All files Region of residence STRATUM/PSU (design info for correct variance estimates) Person and Sample Adult files Education, marital status, sex All files Region of residence STRATUM/PSU (design info for correct variance estimates)

27 Creating the File Data available at the NHIS URL: http://www.cdc.gov/nchs/nhis.htm http://www.cdc.gov/nchs/nhis.htm Data available at the NHIS URL: http://www.cdc.gov/nchs/nhis.htm http://www.cdc.gov/nchs/nhis.htm SAS and SPSS programs are also available to create datasets from the provided data

28 Creating the File Merge the Person, Sample Adult, and Family files together to create one data file Needed to merge files to analyze the association between health insurance coverage and flu shot use Merge the Person, Sample Adult, and Family files together to create one data file Needed to merge files to analyze the association between health insurance coverage and flu shot use

29 Creating the File Each person and each family has a unique identifier (ID) in the NHIS These IDs are used to merge the data sets together Each person and each family has a unique identifier (ID) in the NHIS These IDs are used to merge the data sets together

30 Creating the File Person-level ID Created from household number (HHX) and person number (PX) Family-level ID Created from household number (HHX) and family number (FMX) Person-level ID Created from household number (HHX) and person number (PX) Family-level ID Created from household number (HHX) and family number (FMX)

31 Creating the File Sample Adult file Person file = Adults aged < 65, non-Sample Adults aged > 65, and all children Family file = New file

32 Creating the File Why not drop the records for all children, all Adults aged 65 who were non- Sample Adults? Depending on the situation, this could alter the variance estimates Why not drop the records for all children, all Adults aged 65 who were non- Sample Adults? Depending on the situation, this could alter the variance estimates

33 Creating the File Important to retain the file with all the observations and target the analysis to the particular domain of interest Several software packages for analyzing survey data (such as SUDAAN and STATA) have this capability Important to retain the file with all the observations and target the analysis to the particular domain of interest Several software packages for analyzing survey data (such as SUDAAN and STATA) have this capability

34 Analysis Crosstabs of flu shot propensity among adults > 65 years of age Multiple logistic regression Data from the NHIS 2000 public use files Crosstabs of flu shot propensity among adults > 65 years of age Multiple logistic regression Data from the NHIS 2000 public use files

35 Subpopulation Analyzed 6,180 Sample Adults > 65 years of age Representing a population of 32.7 million 6,180 Sample Adults > 65 years of age Representing a population of 32.7 million

36 Analysis 89 adults > 65 years of age (1%) did not provide their flu shot status and were excluded from the analysis

37 Flu Shot Rates By Health Insurance (aged > 65) Medicaid and Medicare54% Medicare58% Medicare and Private 69% Medicare and other72% Medicaid and Medicare54% Medicare58% Medicare and Private 69% Medicare and other72%

38 Flu Shot Rates By Race/ethnicity (aged > 65) Non-Hispanic black48% Hispanic56% Non-Hispanic other62% Non-Hispanic white67% Non-Hispanic black48% Hispanic56% Non-Hispanic other62% Non-Hispanic white67%

39 Flu Shot Rates By Education (aged > 65) < High School58% High school/GED65% Some college66% A.A. degree66% Bachelor’s degree +74% < High School58% High school/GED65% Some college66% A.A. degree66% Bachelor’s degree +74%

40 Flu Shot Rates By Regular Place of Health Care (aged > 65) Yes65% No25% Yes65% No25%

41 Flu Shot Rates By No. of Physician Office Visits, Last Year (aged > 65) None38% 1 visit60% 2-3 visits61% 4-5 visits67% 6-7 visits69% None38% 1 visit60% 2-3 visits61% 4-5 visits67% 6-7 visits69% 8-9 visits72% 10-12 visits73% 13-15 visits74% 16+ visits75%

42 Odds Ratio (OR) From Logistic Regression dependent variable = flu shot in last 12 months p<0.05 IND. VAR.LEVELORP-VALUE HEALTH INSURANCE Medicare and Medicaid 0.960.74 Medicare(1.00)- Medicare and Private 1.340.00 Medicare and Other 1.890.01

43 Odds Ratio (OR) From Logistic Regression dependent variable = flu shot in last 12 months p<0.05 IND. VAR.LEVELORP-VALUE RACE/ ETHNICITY NH black(1.00)- Hispanic1.400.03 NH other 1.200.53 NH white1.650.00

44 Odds Ratio (OR) From Logistic Regression IND. VAR.LEVELORP-VALUE EDUCATION< HS(1.00)- HS / GED 1.260.00 Some college 1.220.06 A.A. degree 1.300.05 Bachelors + 1.640.00 dependent variable = flu shot in last 12 months p<0.05

45 Odds Ratio (OR) From Logistic Regression dependent variable = flu shot in last 12 months p<0.05 IND. VAR.LEVELORP-VALUE DR. VISITSNone (1.00)- 1 1.970.00 2-3 1.860.00 4-5 2.560.00 6-7 2.720.00 8-9 3.160.00 10-12 3.290.00 13-15 3.290.00 16+ 3.340.00

46 Odds Ratio (OR) From Logistic Regression dependent variable = flu shot in last 12 months p<0.05 IND. VAR.LEVELORP-VALUE REGULAR PLACE OF CARE Yes 2.870.00 No (1.00)-

47 Odds Ratio (OR) From Logistic Regression dependent variable = flu shot in last 12 months p<0.05 IND. VAR.LEVELORP-VALUE REGIONNortheast (1.00)- Midwest 1.010.92 South 1.050.58 West 1.340.01

48 Odds Ratio (OR) From Logistic Regression dependent variable = flu shot in last 12 months p<0.05 IND. VAR.LEVELORP-VALUE MARITAL STATUS Never Married 1.330.28 Married 1.280.04 Separated 1.160.58 Widowed 1.300.03 Divorced(1.00)-

49 Odds Ratio (OR) From Logistic Regression dependent variable = flu shot in last 12 months p<0.05 IND. VAR.LEVELORP-VALUE SEX Female0.890.09 Male(1.00)-

50 Odds Ratio (OR) From Logistic Regression dependent variable = flu shot in last 12 months p<0.05 IND. VAR.LEVELORP-VALUE LOW INCOME PROGRAMS Yes(1.00)- No1.140.25

51 Odds Ratio (OR) From Logistic Regression dependent variable = flu shot in last 12 months p<0.05 IND. VAR.LEVELORP-VALUE SMOKING STATUS Current(1.00)- Former 1.480.00 Never 1.470.00

52 Odds Ratio (OR) From Logistic Regression dependent variable = flu shot in last 12 months p<0.05 IND. VAR.LEVELORP-VALUE HEART DISEASE Yes 1.290.00 No (1.00)-

53 Odds Ratio (OR) From Logistic Regression dependent variable = flu shot in last 12 months p<0.05 IND. VAR.LEVELORP-VALUE DIABETESYes1.010.95 No(1.00)-

54 Odds Ratio (OR) From Logistic Regression dependent variable = flu shot in last 12 months p<0.05 IND. VAR.LEVELORP-VALUE RESP. PROBLEMS Yes1.230.02 No (1.00)-


Download ppt "Analytical Example Using NHIS Data Files John R. Pleis."

Similar presentations


Ads by Google