Presentation is loading. Please wait.

Presentation is loading. Please wait.

CREATE Biostatistics Core THRio Statistical Considerations Analysis of baseline data—esp. truncation Analysis of main study data—esp. correlation.

Similar presentations


Presentation on theme: "CREATE Biostatistics Core THRio Statistical Considerations Analysis of baseline data—esp. truncation Analysis of main study data—esp. correlation."— Presentation transcript:

1 CREATE Biostatistics Core THRio Statistical Considerations Analysis of baseline data—esp. truncation Analysis of main study data—esp. correlation

2 2 Baseline analysis Outcome =TB diagnosis in baseline follow-up period Primary Exposure =1. No HAART & No IPT 2. HAART 3. IPT 4. Both HAART & IPT Sept 1 2003 Sept 1 2005 Outcome =TB diagnosis in baseline follow-up period Primary Exposure =1. No HAART & No IPT 2. HAART 3. IPT 4. Both HAART & IPT Sept 1 2003 Sept 1 2005 Outcome =TB diagnosis in baseline follow-up period Primary Exposure =1. No HAART & No IPT 2. HAART 3. IPT 4. Both HAART & IPT Sept 1 2003 Sept 1 2005

3 3 Study definitions Start = Sept 1, 2003 or HIV diagnosis date if between Sept 1, 2003 and Sept 1, 2005 End = TB diagnosis date or Sept 1, 2005 IPT date = Date that IPT began HAART date =Date that HAART began HIV dx date =Earliest of HIV diagnosis date, initial CD4 date, HAART start date TB dx date = Date that tuberculosis diagnosis reported

4 4 Table 3: Incidence Rate by exposure category Exposure category Person- Years TB cases IR (per 100 PYs) Naïve3,865157 4.06 (3.45-4.75) HAART only11,629229 1.97 (1.72-2.24) IPT only3955 1.27 (0.41-2.95) Both1,25313 1.04 (0.55-1.78) TOTAL17,142404 2.36 (2.13-2.60)

5 5 THRio Baseline Analysis Question: How much should we worry about bias due to truncation / prevalent cohort? --Sickest, by defn, will die earlier. Had to have made at least one visit to a clinic between 1 Sept 2003 and 1 Sept 2005. Not included if died before 1 Sept 2003. Also, someone who died in Nov 2003 would have had little chance to be included.

6 6 truncation… I’m thinking this is somewhat mitigated by controlling for CD4/VL. --Like lining up an analysis of time from HIV seroconversion to TB by estimated conversion time, but staggering entry into risk set according to when came into the study. We have 95% of data on month of first HIV dx. But data would get ‘thin’ if staggered!

7 7 Thinness = entry, at risk Calendar timeline Time since HIV Dx timeline 1 Sept 2003

8 8 Handling Correlation Currently, plan to form daily risk sets, do conditional logistic regression, with a dummy variable for whether each of the 29 clinics is in intervention status on that day (same as Cox model to TB) Correlation can be handled with a sandwich covariance estimator; or, by bootstrapping entire clinic histories Q: sandwich not a great idea when have lots of obs per cluster and few clusters; but what if those lots of obs only have a few events? Perhaps 10-20 TB events per clinic.


Download ppt "CREATE Biostatistics Core THRio Statistical Considerations Analysis of baseline data—esp. truncation Analysis of main study data—esp. correlation."

Similar presentations


Ads by Google