Skip to main content
. Author manuscript; available in PMC: 2020 Apr 1.
Published in final edited form as: Am J Epidemiol. 2019 Apr 1;188(4):709–723. doi: 10.1093/aje/kwy265

Table 3.

Empirical Example 1: Results for Time-to-Event Outcomes from Propensity Score-Adjusted Analyses using Different Combinations of Confounding Adjustment Method and Data-Sharing Approach, AGB vs RYGBa

Confounding adjustment method & data-sharing approach Effectiveness outcome, change in body mass indexb Safety outcomec
<5% ≥5% ≥10% ≥20% ≥30% Re-hospitalization Re-intervention
HR 95% CI HR 95% CI HR 95% CI HR 95% CI HR 95% CI HR 95% CI HR 95% CI
Stratification
 Pooled individual-level 2.20 1.97, 2.46 0.53 0.49, 0.58 0.36 0.32, 0.40 0.17 0.15, 0.20 0.10 0.07, 0.13 0.84 0.64, 1.12 0.82 0.61, 1.09
 Risk-set 2.20 1.97, 2.46 0.53 0.49, 0.58 0.36 0.32, 0.40 0.17 0.15, 0.20 0.10 0.07, 0.13 0.84 0.64, 1.12 0.82 0.61, 1.09
 Summary-table 3.48 3.11, 3.89 0.42 0.39, 0.46d 0.37 0.34, 0.41d 0.22 0.19, 0.26d 0.12 0.08, 0.17 0.84 0.62, 1.11 0.81 0.60, 1.09
 Effect-estimate, fixed-effect 2.26 2.02, 2.52 0.54 0.49, 0.59 0.36 0.33, 0.40 0.17 0.15, 0.20 0.11 0.08, 0.15 0.97 0.73, 1.28 0.85 0.64, 1.14
 Effect-estimate, random-effects 2.35 1.21, 4.56 0.58 0.41, 0.81 0.36 0.27, 0.47 0.15 0.10, 0.22 0.06 0.02, 0.19 0.85 0.32, 2.22 0.76 0.45, 1.28
 Measure of heterogeneity, Qe 25.29 <.0001 15.08 0.0005 7.80 0.0202 6.55 0.0377 6.66 0.0357 13.50 0.0012 3.97 0.1371
Matching
 Pooled individual-level 2.25 1.91, 2.66 0.50 0.44, 0.57 0.35 0.31, 0.40 0.17 0.15, 0.21 0.10 0.07, 0.14 0.85 0.60, 1.22 0.78 0.54, 1.11
 Risk-set 2.25 1.91, 2.66 0.50 0.44, 0.57 0.35 0.31, 0.40 0.17 0.15, 0.21 0.10 0.07, 0.14 0.85 0.60, 1.22 0.78 0.54, 1.11
 Summary-table 3.49 2.95, 4.13 0.40 0.36, 0.45 0.38 0.33, 0.43 0.23 0.19, 0.27 0.12 0.09, 0.17 0.85 0.58, 1.23 0.77 0.52, 1.12
 Effect-estimate, fixed-effect 2.23 1.89, 2.63 0.50 0.45, 0.57 0.35 0.31, 0.40 0.17 0.15, 0.21 0.11 0.08, 0.15 0.91 0.63, 1.32 0.81 0.56, 1.18
 Effect-estimate, random-effects 1.95 0.92, 4.16 0.56 0.32, 0.99 0.36 0.27, 0.47 0.17 0.12, 0.23 0.07 0.02, 0.23 0.73 0.29, 1.87 0.64 0.24, 1.72
 Measure of heterogeneity, Qe 15.58 0.0004 22.17 <.0001 4.79 0.0911 3.24 0.1976 5.58 0.0612 7.65 0.0218 7.71 0.0211
Inverse probability weighting
 Pooled individual-level 1.92 1.73, 2.13 0.49 0.45, 0.54 0.32 0.29, 0.35 0.19 0.17, 0.22 0.10 0.08, 0.14 0.84 0.65, 1.08 0.85 0.65, 1.12
 Risk-set 1.93 1.51, 2.47 0.49 0.40, 0.59 0.32 0.26, 0.38 0.19 0.14, 0.27 0.10 0.06, 0.17 0.84 0.46, 1.53 0.80 0.60, 1.07
 Effect-estimate, fixed-effect 1.92 1.73, 2.13 0.50 0.46, 0.54 0.32 0.29, 0.35 0.20 0.18, 0.23 0.15 0.11, 0.20 1.16 0.89, 1.51 1.07 0.81, 1.40
 Effect-estimate, random-effects 1.85 1.55, 2.21 0.54 0.43, 0.68 0.32 0.25, 0.41 0.19 0.11, 0.33 0.02 0.00, 0.42 0.74 0.22, 2.51 0.45 0.11, 1.80
 Measure of heterogeneity, Qe 3.31 0.1905 9.67 0.0079 9.24 0.0098 22.16 <.0001 10.49 0.0053 27.68 <.0001 15.27 0.0005
Matching weighting
 Pooled individual-level 2.29 1.94, 2.71 0.53 0.46, 0.59 0.34 0.30, 0.39 0.17 0.14, 0.20 0.10 0.07, 0.14 0.80 0.56, 1.15 0.83 0.57, 1.20
 Risk-set 2.33 2.08, 2.62 0.52 0.47, 0.57 0.34 0.31, 0.38 0.17 0.14, 0.20 0.10 0.07, 0.13 0.80 0.60, 1.07 0.85 0.57, 1.28
 Effect-estimate, fixed-effect 2.26 1.91, 2.67 0.52 0.46, 0.59 0.34 0.30, 0.39 0.17 0.14, 0.20 0.11 0.08, 0.15 0.83 0.57, 1.20 0.83 0.57, 1.22
 Effect-estimate, random-effects 2.23 1.11, 4.49 0.56 0.40, 0.78 0.34 0.26, 0.44 0.14 0.09, 0.23 0.06 0.02, 0.20 0.77 0.27, 2.24 0.81 0.53, 1.25
 Measure of heterogeneity, Qe 10.84 0.0044 7.90 0.0192 4.05 0.1316 6.28 <.0431 6.20 0.0450 8.35 0.0154 2.22 0.3285

Note: AGB= Adjusted gastric banding; RYGB= Roux-en-Y gastric bypass; HR= hazard ratio; CI= confidence interval

a

There were 584 (6.2%) patients who underwent AGB and 8,777 (93.8%) patients who underwent RYGB.

b

The incidences for <5%, ≥5%, ≥10%, ≥20%, and ≥30% change in body mass index were 68.0%, 93.7%, 76.5%, 31.2%, and 6.8%, respectively, for the AGB users; 32.9%, 99.1%, 96.8%, 83.8%, and 47.7%, respectively, for the RYGB users. These effectiveness outcomes were defined as the occurrence of the outcomes of interest closet to the end of the first post-procedure year so the incidences for <5% change in body mass index and ≥5% change in body mass index do not sum up to 100%.

c

The incidences for re-hospitalization and re-intervention were 9.9% and 9.3%, respectively, for the AGB users, and 11.7% and 11.6%, respectively, for the RYGB users.

d

These were calculated using the Mantel-Haenszel approach, because the exact confidence intervals for the regression-based analysis could not be obtained.

e

Q is a measure of heterogeneity among the three data-contributing sites. The summary statistic and p-value from Cochran’s Q test are shown here.