Common use of Statistical Analysis Clause in Contracts

Statistical Analysis. The research questions of interested examined the relationship between the two area- level exposures, area disadvantage and local racial/ethnic spatial concentration, and HIV testing. We first describe the study population by their individual- and ZCTA-level characteristics by area disadvantage and local racial/ethnic spatial concentration and present the prevalence of ever having tested for HIV stratified by the disadvantage and ICE quintiles for the three race/ethnic groups. For the multivariable analysis, we conducted multilevel modeling with a modified Poisson approach (111) to estimate prevalence ratios and used generalized estimating equations, with an exchangeable correlation structure and robust variance estimation, to account for clustering by ZCTA. Subjects with missing data for the outcome and/or covariates were excluded from the analyses. Area disadvantage and ICE and their relationships with the outcome were examined separately. Our modeling strategy fit a series of models: (i) unadjusted without interaction, (ii) exposure X race interaction-only model, (iii) fully adjusted model adjusting for individual- and area-level covariates with exposure x race interaction. For the models with interaction, we assessed multiplicative interaction with the generalized score test statistic and its p-value and additive interaction with the relative excess risk due to interaction (RERI) from the relative risk model (112). A RERI calculated using risk ratios (RERIRR) indicates the direction of the additive interaction, whether positive or negative, but not the relative magnitude of the interaction. For confounding, we adjusted for individual- and area-level covariates, based on prior literature (21,22,27,35,124): age; sexual identity; educational status, whether on track or not based on age; sexual identity disclosure; experiences of sexual identity- related stigma; health insurance; health care provider visit in the past one year; STI testing and diagnosis in the past year; condomless anal intercourse; urban-rural residence; region of the country; density of HIV testing sites within the ZIP Code; and distance to HIV testing. We also examined the predicted probabilities of ever testing at each exposure quintile by race/ethnicity based on parameters for the fully adjusted model holding all covariates constant at their average. Multicollinearity was assessed by examining condition indices (greater than 35) and variance decomposition factors (two or more greater than 0.5). Since these criteria were not met, all covariates were retained in the final model. The findings presented focus on the comparisons between the extremes - the most (Q5) and least (Q1) disadvantaged ZCTAs and the highest POC concentration (Q5) and highest White concentration ZCTAs (Q1). We did not include Asian and Pacific Islander participants and participants who reported their race/ethnicity as Other or Multiracial in the regression analyses. The small sample size of Asian and Pacific Islander participants would produce unreliable estimates, and the heterogeneity of the Other/Multiracial category precluded us from making meaningful interpretations about the exposure-outcome relationships for this group. Descriptive and regression analyses were conducted using SAS 9.4 (Xxxxx, NC, USA). Calculation of distance to the nearest HIV testing site was conducted using R 3.6.2 (Vienna, Austria). AMIS was reviewed and approved by the Emory University Institutional Review Board. RESULTS

Statistical Analysis. The research questions of interested examined the relationship between the two area- level exposures, area disadvantage and local racial/ethnic spatial concentration, and HIV testing. We first describe the study population by their individual- and ZCTA-level characteristics by area disadvantage and local racial/ethnic spatial concentration and present the prevalence of ever having tested for HIV stratified by the disadvantage and ICE quintiles for the three race/ethnic groups. For the multivariable analysis, we conducted multilevel modeling with a modified Poisson approach (111) to estimate prevalence ratios and used generalized estimating equations, equations with an exchangeable correlation structure and robust variance estimation, to account for clustering by ZCTA. Subjects with missing data for the outcome and/or covariates were excluded from the analyses. Area disadvantage and ICE and their relationships with the outcome were examined separately. Our modeling strategy fit a series of models: (i) unadjusted without interaction, (ii) exposure X race interaction-only model, (iii) fully adjusted model adjusting for individual- and area-level covariates with exposure x race interaction. For the models with interaction, we assessed multiplicative interaction with the generalized score test statistic and its p-value and additive interaction with the relative excess risk due to interaction (RERI) from the relative risk model (112). A RERI calculated using risk ratios (RERIRR) indicates the direction of the additive interaction, whether positive or negative, but not the relative magnitude of the interaction. For To address confounding, we adjusted for the following individual- and area-level covariates, covariates based on prior literature (21,22,27,35,12435,41,97,113): ageage in years; sexual identity; educational status, whether on track or not based on age; sexual identity disclosure; experiences of sexual identity- related stigma; health insurance; health care provider visit in the past one year; CAI; STI testing and diagnosis in the past year; condomless anal intercoursehaving ever tested for HIV; urban-having heard of PrEP; urban- rural residence; and region of the country; density of HIV testing sites within the ZIP Code; and distance to HIV testing. We also examined the predicted probabilities of ever testing at each exposure quintile by race/ethnicity based on parameters for the fully adjusted model holding all covariates constant at their average. Multicollinearity was assessed by examining condition indices (greater than 35) and variance decomposition factors (two or more greater than 0.5). Since these criteria were not met, all covariates were retained in the final model. The findings presented focus on the comparisons between the extremes - the most (Q5) and least (Q1) disadvantaged ZCTAs and the highest POC concentration (Q5) and highest White concentration ZCTAs (Q1). We did not include Asian and Pacific Islander participants and participants who reported their race/ethnicity as Other or Multiracial in the regression analyses. The small sample size of Asian and Pacific Islander participants would produce unreliable estimates, and the heterogeneity of the Other/Multiracial category precluded us from making meaningful interpretations about the exposure-outcome relationships for this group. Descriptive and regression analyses were conducted using SAS 9.4 (XxxxxCarey, NC, USA). Calculation of distance to the nearest HIV testing site was conducted using R 3.6.2 (Vienna, Austria). AMIS XXXX was reviewed and approved by the Emory University Institutional Review Board. RESULTS

Statistical Analysis. The research questions of interested examined the relationship between the two area- level exposures, area disadvantage and local racial/ethnic spatial concentration, and HIV testing. We first describe the study population by their individual- and ZCTA-level characteristics by area disadvantage and local racial/ethnic spatial concentration and present the prevalence of ever having tested for HIV stratified by the disadvantage and ICE quintiles for the three race/ethnic groups. For the multivariable analysis, we conducted multilevel modeling with a modified Poisson approach (111) to estimate prevalence ratios and used generalized estimating equations, with an exchangeable correlation structure and robust variance estimation, to account for clustering by ZCTA. Subjects with missing data for the outcome and/or covariates were excluded from the analyses. Area disadvantage and ICE and their relationships with the outcome were examined separately. Our modeling strategy fit a series of models: (i) unadjusted without interaction, (ii) exposure X race interaction-only model, (iii) fully adjusted model adjusting for individual- and area-level covariates with exposure x race interaction. For the models with interaction, we assessed multiplicative interaction with the generalized score test statistic and its p-value and additive interaction with the relative excess risk due to interaction (RERI) from the relative risk model (112). A RERI calculated using risk ratios (RERIRR) indicates the direction of the additive interaction, whether positive or negative, but not the relative magnitude of the interaction. For confounding, we adjusted for individual- and area-level covariates, based on prior literature (21,22,27,35,124): age; sexual identity; educational status, whether on track or not based on age; sexual identity disclosure; experiences of sexual identity- related stigma; health insurance; health care provider visit in the past one year; STI testing and diagnosis in the past year; condomless anal intercourse; urban-rural residence; region of the country; density of HIV testing sites within the ZIP Code; and distance to HIV testing. We also examined the predicted probabilities of ever testing at each exposure quintile by race/ethnicity based on parameters for the fully adjusted model holding all covariates constant at their average. Multicollinearity was assessed by examining condition indices (greater than 35) and variance decomposition factors (two or more greater than 0.5). Since these criteria were not met, all covariates were retained in the final model. The findings presented focus on the comparisons between the extremes - the most (Q5) and least (Q1) disadvantaged ZCTAs and the highest POC concentration (Q5) and highest White concentration ZCTAs (Q1). We did not include Asian and Pacific Islander participants and participants who reported their race/ethnicity as Other or Multiracial in the regression analyses. The small sample size of Asian and Pacific Islander participants would produce unreliable estimates, and the heterogeneity of the Other/Multiracial category precluded us from making meaningful interpretations about the exposure-outcome relationships for this group. Descriptive and regression analyses were conducted using SAS 9.4 (XxxxxCarey, NC, USA). Calculation of distance to the nearest HIV testing site was conducted using R 3.6.2 (Vienna, Austria). AMIS XXXX was reviewed and approved by the Emory University Institutional Review Board. RESULTS

Statistical Analysis. The research questions of interested examined the relationship between the two area- level exposures, area disadvantage and local racial/ethnic spatial concentration, and HIV testing. We first describe the study population by their individual- and ZCTA-level characteristics by area disadvantage and local racial/ethnic spatial concentration and present the prevalence of ever having tested for HIV stratified by the disadvantage and ICE quintiles for the three race/ethnic groups. For the multivariable analysis, we conducted multilevel modeling with a modified Poisson approach (111) to estimate prevalence ratios and used generalized estimating equations, equations with an exchangeable correlation structure and robust variance estimation, to account for clustering by ZCTA. Subjects with missing data for the outcome and/or covariates were excluded from the analyses. Area disadvantage and ICE and their relationships with the outcome were examined separately. Our modeling strategy fit a series of models: (i) unadjusted without interaction, (ii) exposure X race interaction-only model, (iii) fully adjusted model adjusting for individual- and area-level covariates with exposure x race interaction. For the models with interaction, we assessed multiplicative interaction with the generalized score test statistic and its p-value and additive interaction with the relative excess risk due to interaction (RERI) from the relative risk model (112). A RERI calculated using risk ratios (RERIRR) indicates the direction of the additive interaction, whether positive or negative, but not the relative magnitude of the interaction. For To address confounding, we adjusted for the following individual- and area-level covariates, covariates based on prior literature (21,22,27,35,12435,41,97,113): ageage in years; sexual identity; educational status, whether on track or not based on age; sexual identity disclosure; experiences of sexual identity- related stigma; health insurance; health care provider visit in the past one year; CAI; STI testing and diagnosis in the past year; condomless anal intercoursehaving ever tested for HIV; urban-having heard of PrEP; urban- rural residence; and region of the country; density of HIV testing sites within the ZIP Code; and distance to HIV testing. We also examined the predicted probabilities of ever testing at each exposure quintile by race/ethnicity based on parameters for the fully adjusted model holding all covariates constant at their average. Multicollinearity was assessed by examining condition indices (greater than 35) and variance decomposition factors (two or more greater than 0.5). Since these criteria were not met, all covariates were retained in the final model. The findings presented focus on the comparisons between the extremes - the most (Q5) and least (Q1) disadvantaged ZCTAs and the highest POC concentration (Q5) and highest White concentration ZCTAs (Q1). We did not include Asian and Pacific Islander participants and participants who reported their race/ethnicity as Other or Multiracial in the regression analyses. The small sample size of Asian and Pacific Islander participants would produce unreliable estimates, and the heterogeneity of the Other/Multiracial category precluded us from making meaningful interpretations about the exposure-outcome relationships for this group. Descriptive and regression analyses were conducted using SAS 9.4 (Xxxxx, NC, USA). Calculation of distance to the nearest HIV testing site was conducted using R 3.6.2 (Vienna, Austria). AMIS was reviewed and approved by the Emory University Institutional Review Board. RESULTS