Statistical analysis
Stata (Version Intercooled 8.0, College Station, TX) was used for all statistical analyses. Analysis of study data was carried out in a two-stage process. First, general estimates of anemia prevalence, hemoglobin level, and its association with key resident characteristics were explored with cross-tabulations and univariate statistical analysis. Secondly, to control for underlying differences in key characteristics among cohorts in this retrospective analysis, several multivariate analyses were conducted to further explore these relationships.
Patients missing data for the critical variables sex, race or index Hb level were removed from analysis. Since the multivariate models employed required complete data for all retained residents, values for missing data in other variables analyzed in this study were imputed from a multiple linear regression model (Stata Impute procedure) using 22 baseline resident characteristics.
A multiple regression model was used to analyze the relationship between selected resident characteristics and comorbidities with index Hb level. Diagnostics (examination of variance inflation factors and the model coefficient correlation matrix) were performed to evaluate potential multicollinearity. Logistic regression was then used to analyze the relationship between both falling and recurrent falls with anemia (using the gender-specific WHO definition), adjusting for selected variables presumed to be related to falls. All incidents of falls for the 6-month post-index period were divided into two measures: whether the resident experienced falling (had one or more falls) or whether the resident had recurrent falls (had two or more falls). Additionally, to explore whether index Hb level had a linear relationship with falls, the same models were replicated by substituting index Hb level and gender in place of anemia status. Starting with the full model (see Table 1 for a listing of included variables), stepwise backward elimination (maximum likelihood method) was used to progressively remove one variable at a time from the model until the p-value (Wald statistic) of any coefficient no longer exceeded 0.25. Results for the final, reduced models were retained.
For a multiple regression model, the required sample size is estimated at 178 (¼0.05, 27 predictors, medium effect size f 2¼0.15 and power of 0.8). For the logistical regression models, conservatively assuming that falls would occur for only 5% of residents over the observation period, the required sample size is 460 (¼0.05, r¼0.5 between true and measured exposure, odds ratio for one SD increase in covariate¼1.5, odds ratio for one SD increase in true exposure¼2.0, r¼0 between exposure and covariate).
Statistical analysis
Stata (Version Intercooled 8.0, College Station, TX) was used for all statistical analyses. Analysis of study data was carried out in a two-stage process. First, general estimates of anemia prevalence, hemoglobin level, and its association with key resident characteristics were explored with cross-tabulations and univariate statistical analysis. Secondly, to control for underlying differences in key characteristics among cohorts in this retrospective analysis, several multivariate analyses were conducted to further explore these relationships.
Patients missing data for the critical variables sex, race or index Hb level were removed from analysis. Since the multivariate models employed required complete data for all retained residents, values for missing data in other variables analyzed in this study were imputed from a multiple linear regression model (Stata Impute procedure) using 22 baseline resident characteristics.
A multiple regression model was used to analyze the relationship between selected resident characteristics and comorbidities with index Hb level. Diagnostics (examination of variance inflation factors and the model coefficient correlation matrix) were performed to evaluate potential multicollinearity. Logistic regression was then used to analyze the relationship between both falling and recurrent falls with anemia (using the gender-specific WHO definition), adjusting for selected variables presumed to be related to falls. All incidents of falls for the 6-month post-index period were divided into two measures: whether the resident experienced falling (had one or more falls) or whether the resident had recurrent falls (had two or more falls). Additionally, to explore whether index Hb level had a linear relationship with falls, the same models were replicated by substituting index Hb level and gender in place of anemia status. Starting with the full model (see Table 1 for a listing of included variables), stepwise backward elimination (maximum likelihood method) was used to progressively remove one variable at a time from the model until the p-value (Wald statistic) of any coefficient no longer exceeded 0.25. Results for the final, reduced models were retained.
For a multiple regression model, the required sample size is estimated at 178 (¼0.05, 27 predictors, medium effect size f 2¼0.15 and power of 0.8). For the logistical regression models, conservatively assuming that falls would occur for only 5% of residents over the observation period, the required sample size is 460 (¼0.05, r¼0.5 between true and measured exposure, odds ratio for one SD increase in covariate¼1.5, odds ratio for one SD increase in true exposure¼2.0, r¼0 between exposure and covariate).
การแปล กรุณารอสักครู่..

การวิเคราะห์ทางสถิติ ( Language
intercooled วิทยาลัยรุ่น 8.0 , สถานี , TX ) สถิติที่ใช้ในการวิเคราะห์ทางสถิติทั้งหมด การวิเคราะห์ข้อมูลได้ดำเนินการในกระบวนการสองขั้นตอน . แรก , ประมาณการทั่วไปโลหิตจางชุก ระดับฮีโมโกลบิน และความสัมพันธ์กับลักษณะสำคัญคือ tabulations ข้ามถิ่นขั้นพื้นฐานและการวิเคราะห์ทางสถิติ 2 . ประการที่สอง to control for underlying differences in key characteristics among cohorts in this retrospective analysis, several multivariate analyses were conducted to further explore these relationships.
Patients missing data for the critical variables sex, race or index Hb level were removed from analysis. Since the multivariate models employed required complete data for all retained residents, values for missing data in other variables analyzed in this study were imputed from a multiple linear regression model (Stata Impute procedure) using 22 baseline resident characteristics.
A multiple regression model was used to analyze the relationship between selected resident characteristics and comorbidities with index Hb level. Diagnostics (examination of variance inflation factors and the model coefficient correlation matrix) were performed to evaluate potential multicollinearity. Logistic regression was then used to analyze the relationship between both falling and recurrent falls with anemia (using the gender-specific WHO definition), adjusting for selected variables presumed to be related to falls.เหตุการณ์ทั้งหมดก็ 6 เดือนหลังดัชนีระยะเวลาแบ่งออกเป็น 2 มาตรการว่า ถิ่นที่มีประสบการณ์ลดลง ( มีหนึ่งหรือมากกว่าหนึ่งตก ) หรือว่ามีถิ่นที่อยู่ได้ตกซ้ำ ( มีสองคนหรือมากกว่าอยู่ ) นอกจากนี้ เพื่อสำรวจว่าระดับดัชนีมีความสัมพันธ์เชิงเส้นกับ HB ตก the same models were replicated by substituting index Hb level and gender in place of anemia status. Starting with the full model (see Table 1 for a listing of included variables), stepwise backward elimination (maximum likelihood method) was used to progressively remove one variable at a time from the model until the p-value (Wald statistic) of any coefficient no longer exceeded 0.25.ผลการค้นหาสำหรับสุดท้ายลดแบบถูกเก็บไว้ .
สำหรับการวิเคราะห์ถดถอยพหุคูณแบบ กำหนดขนาดตัวอย่างประมาณ 178 ( ¼ 0.05 27 ตัว ขนาดกลาง ผลขนาด F 2 ¼ 0.15 และพลังของ 0.8 ) สำหรับตัวแบบการถดถอยโลจิสติคส์ อนุรักษ์นิยมคิดว่าตกจะเกิดขึ้นเพียง 5% ของประชากรในช่วงเวลาการสังเกต ต้องใช้ขนาดตัวอย่าง 460 ( ¼ 0.05 R ¼ 05 between true and measured exposure, odds ratio for one SD increase in covariate¼1.5, odds ratio for one SD increase in true exposure¼2.0, r¼0 between exposure and covariate).
การแปล กรุณารอสักครู่..
