Comparisons of electronically delivered and in-person cognitive–behavioral therapy (CBT) become difficult when patients choose their delivery mode. Pretreatment severity may influence that choice, and incomplete follow-up may conceal clinically important differences even when conventional interaction tests are not statistically significant. This study asked how much residual preference-related distortion would be required to overturn an endpoint conclusion of comparable proportional improvement, and whether completion and persistence evidence supported the same interpretation. A reproducible aggregate cohort was constructed from reported arm-level summaries for 108 adults with major depressive disorder who selected therapist-supported electronic CBT (e-CBT; \(n=53\)) or in-person CBT (\(n=55\)). Preference–Attrition Tipping-Point Analysis (PATPA) combined four elements: Hedges-corrected pretreatment contrasts, Jeffreys-prior beta–binomial completion estimates, a conditional analysis of sessions completed before withdrawal, and a rounding-aware bias boundary for normalized Patient Health Questionnaire-9 (PHQ-9) change. A 10-percentage-point region was specified for completion and normalized endpoint contrasts. Pretreatment symptom differences were substantial: Hedges’ \(g\) for e-CBT minus in-person CBT was \(-0.72\) for PHQ-9 and \(-0.97\) for the Quick Inventory of Depressive Symptomatology–Self Report; quality of life favored e-CBT (\(g=0.45\)). Completion proportions were nearly identical (45.3% and 45.5%), but the posterior 95% credible interval for their difference was wide (\(-18.5\) to \(18.2\) percentage points), and the probability of lying within the prespecified 10-point region was 0.710. Among participants who withdrew, e-CBT was associated with 1.65 additional sessions before withdrawal (95% interval 0.40–2.90). Normalized PHQ-9 values at week 12 were 0.78 and 0.79, producing an endpoint contrast of \(-0.01\). The comparability conclusion held only while residual preference bias remained between \(-0.11\) and \(0.09\) of the pretreatment level. Apparent endpoint similarity was compatible with modest residual preference distortion, but it was not immune to it. Completion equality was imprecisely estimated, whereas persistence among withdrawals favored e-CBT conditionally. Delivery-mode conclusions should therefore report a bias boundary and attrition uncertainty rather than rely on a non-significant treatment-by-time interaction.
Comparisons of electronically delivered and in-person cognitive–behavioral therapy (CBT) become difficult when patients choose their delivery mode. Pretreatment severity may influence that choice, and incomplete follow-up may conceal clinically important differences even when conventional interaction tests are not statistically significant. This study asked how much residual preference-related distortion would be required to overturn an endpoint conclusion of comparable proportional improvement, and whether completion and persistence evidence supported the same interpretation. A reproducible aggregate cohort was constructed from reported arm-level summaries for 108 adults with major depressive disorder who selected therapist-supported electronic CBT (e-CBT; \(n=53\)) or in-person CBT (\(n=55\)). Preference–Attrition Tipping-Point Analysis (PATPA) combined four elements: Hedges-corrected pretreatment contrasts, Jeffreys-prior beta–binomial completion estimates, a conditional analysis of sessions completed before withdrawal, and a rounding-aware bias boundary for normalized Patient Health Questionnaire-9 (PHQ-9) change. A 10-percentage-point region was specified for completion and normalized endpoint contrasts. Pretreatment symptom differences were substantial: Hedges’ \(g\) for e-CBT minus in-person CBT was \(-0.72\) for PHQ-9 and \(-0.97\) for the Quick Inventory of Depressive Symptomatology–Self Report; quality of life favored e-CBT (\(g=0.45\)). Completion proportions were nearly identical (45.3% and 45.5%), but the posterior 95% credible interval for their difference was wide (\(-18.5\) to \(18.2\) percentage points), and the probability of lying within the prespecified 10-point region was 0.710. Among participants who withdrew, e-CBT was associated with 1.65 additional sessions before withdrawal (95% interval 0.40–2.90). Normalized PHQ-9 values at week 12 were 0.78 and 0.79, producing an endpoint contrast of \(-0.01\). The comparability conclusion held only while residual preference bias remained between \(-0.11\) and \(0.09\) of the pretreatment level. Apparent endpoint similarity was compatible with modest residual preference distortion, but it was not immune to it. Completion equality was imprecisely estimated, whereas persistence among withdrawals favored e-CBT conditionally. Delivery-mode conclusions should therefore report a bias boundary and attrition uncertainty rather than rely on a non-significant treatment-by-time interaction.
