Meta-analysis applications Assignment
STATA Exercise Homework
Reminder: you can find helpful information regarding each STATA meta-analysis command by typing “help command” (where command is metan, metareg, metacum, metabias, etc.) or consulting the handout on STATA notes.
Questions 1 to 3 of this STATA exercise homework are based on 21 randomized controlled trials conducted from 1959 to 1986 investigating the effect of streptokinase on lowering mortality from myocardial infarction. Use risk ratios to quantify the association between streptokinase and myocardial infarction. RR = (number of cases among exposed (a)/ total exposed (N_{1}))/(number of cases among non-exposed (b)/total non-exposed (N_{2})). Var(RR) = [(N_{1} – a)/aN_{1}] + [(N_{2} – b)/bN_{2}]
Below is a summary of the variables in the accompanying homework dataset. The dataset, strepto_21trials_v4.dta, can be access on the course website.
{`+---------------------------------------------------------+ | trialnam year pop1 deaths1 pop0 deaths0 | |---------------------------------------------------------| 1. | 1st Australian 1973 264 26 253 32 | 2. | 1st European 1969 83 20 84 15 | 3. | 2nd Australian 1977 112 25 118 31 | 4. | 2nd European 1971 373 69 357 94 | 5. | 2nd Frankfurt 1973 102 13 104 29 | |---------------------------------------------------------| 6. | 3rd European 1977 156 25 159 50 | 7. | Austrian 1977 352 37 376 65 | 8. | Dewar 1963 21 4 21 7 | 9. | Fletcher 1959 12 1 11 4 | 10. | Frank 1975 55 6 53 6 | |---------------------------------------------------------| 11. | GISSI-1 1986 5860 628 5852 758 | 12. | Heikinheimo 1971 219 22 207 17 | 13. | ISAM 1986 859 54 882 63 | 14. | Italian 1971 164 19 157 18 | 15. | Klein 1976 14 4 9 1 | |---------------------------------------------------------| 16. | Lasierra 1977 13 1 11 3 | 17. | N German 1977 249 63 234 51 | 18. | NHLBI SMIT 1974 53 7 54 3 | 19. | UK Collab 1976 302 48 293 52 | 20. | Valere 1975 49 11 42 9 | 21. | Witchitz 1977 32 5 26 5 | |---------------------------------------------------------| `}
Question 1:
- Using the “metan” command with direct input of the 2x2 table cell counts, conduct a fixed-effect inverse variance pooling of the study specific RRs and copy the resulting forest plot. What is the pooled RR and 95% confidence interval? What does this say about the effect of streptokinase on myocardial infarction?
- Is there evidence of significant heterogeneity? Report the p value for heterogeneity, as well as the I-squared statistic and 95% CI of the I^{2}.
- Conduct a random-effect meta-analysis. Report the new pooled RR and 95% confidence interval and copy the resulting forest plot. Comparing the results to the fixed-effect analysis, which confidence interval is larger? Why? Are the % weights the same as in the fixed-effects analysis? Why or why not?
- Using the study-specific logRR and SE(logRR), compute the fixed and random effects summary estimates. How do they compare to the analysis when the numbers of cases and noncases were used?
Question 2
- Use “metacum” to examine the pooled effect estimate by adding one study at a time over time. Is there evidence of changing magnitude in pooled effect estimate over time?
- Did any particular study or studies have a strong influence on the pooled estimate?
- Apply meta-regression analysis “metareg” to examine if there was evidence of effect modification by baseline mortality rate (estimated from the controls) in a trial.
Question 3:
- Use the “metabias” command to examine whether publication bias existed using Egger’s test. Paste your output as well as both the funnel plot and Egger’s regression plot.
- Do the results statistically indicate possible publication bias? If so, which tests?
- Aside from p-values, describe what one visually looks for in a funnel plot and in Egger’s plot to determine whether there is possible publication bias.
Question 4:
The following data are from five fabricated studies of the effect of soy-protein intake on LDL-C reduction. You will find the data file in the “soy_5b.dta” on the website.
+-----------------------------------------------+ | author dose duration soychg soysd soyn conchg consd conn | |--------------------------------------------| 1. | Aggregat 30g/day 2w -75 35 24 -4 5 24 | 2. | Bundled 20g/day 3w -42 35 12 -29 22 12 | 3. | Combini 25g/day 4w -45 23 43 -4 5 44 | 4. | Deposita 15g/day 1y -34 64 22 4 8 21 | 5. | Embodyl 50g/day 6w -24 23 53 -3 6 44 |+--------------------------------------+
Below is a summary of the variables in this homework dataset.
Variable |
Description |
author |
first author of the trial |
dose |
amount of soy protein given in the treatment arm |
duration |
duration of soy intake in the treatment arm |
soychg |
mean LDL-C change (mg/dl) on soy from baseline to end of trial |
soysd |
SD of LDL-C change on soy |
soyn |
number of participants analyzed in the soy arm |
conchg |
mean LDL-C change (mg/dl) on control from baseline to end of trial |
consd |
SD of LDL-C change on control |
conn |
number of participants analyzed in the control arm |
- Calculate the weighted mean difference between the "soy" treatment arms and "control", using the “metan” command. Summarize the effect size in one sentence, and paste you forest plot.
- Calculate the standardized mean difference between the "soy" treatment arms and "control", using the “metan” command. Summarize the effect size in one sentence, and paste you forest plot.
- You have written your manuscript, and are passing it around to your co-authors. One author would like to report the WMD, and another would like to report the SMD. However, they have not taken EPI 233, and defer to your judgment. Which would you prefer to include? Please justify your choice in one or two sentences.
- How would you deal with the heterogeneity in your meta-analysis? Please briefly motivate your choices.