Source-linked reanalysis of psychological treatment for health anxiety

Run in Stata 19 or later, setting this directory as the current directory:
    do run_all.do

The code uses Stata built-in meta commands; no user-written packages are required.
It clears data in the Stata instance that runs it. Save unrelated work first.
The submitted inputs are published aggregate summaries, not participant data.

There are 45 result rows. These are not 45 independent studies. Rows include
repeated times, shared arms, alternative estimands and sensitivity assumptions.
The 28 labels cover 27 retained canonical families and one report whose
independence from another retained report is unresolved. Consult the review's
family crosswalk before interpreting or combining results.

All contrasts are first-listed intervention minus comparator; lower health-
anxiety scores are better. Native exact Hedges g and its Stata default variance
are used for raw standardized endpoint effects. Raw mean-difference intervals
use Welch t. Published confidence bounds remain unaltered for reported models.
Reported estimates, adjusted changes and raw endpoints must not be pooled
indiscriminately. Sample-size fields retain their original analysis meanings.

The four overlapping CBT/waiting-list pools are exploratory methodological
sensitivities, not the complete body of eligible CBT/waiting-list evidence.
They cannot be treated as four independent replications. The largest subset
includes an uncorrected partial-nesting sensitivity. Lee's SD assumption,
LOCF summaries and secondary-scale contrasts are documented in provenance.

source_provenance.csv links inputs to publications and source locations.
Copyrighted full-text reports are not redistributed. Code and extraction were
AI-assisted; arithmetic checks do not establish independent human review.
