BACKGROUND: According to meta-analyses of randomised controlled trials (RCTs), therapist-guided internet-delivered cognitive behavioural therapy (iCBT) is as effective a treatment for depression as traditional face-to-face CBT (fCBT), despite its substantially lower costs. However, RCTs are not always representative of routine practice, which could inflate effectiveness estimates. We combined rich data with counterfactual causal statistical reasoning to provide an fCBT-iCBT comparison complementary to RCTs. METHODS: In this retrospective cohort study, we linked full archived therapist-guided iCBT and fCBT cohort registries with multiple Finnish social and health care registries. The therapist-guided iCBT programme with third-wave principles and the fCBTs were provided by HUS Helsinki University Hospital to people with depression without acute suicide or substance-misuse risk
fCBT was delivered in the Uusimaa region, whereas therapist-guided iCBT was nationwide and excluded people with treatment-interfering psychotic, neurological, or personality disorders, chronic or bipolar depression, or aged under 16 years. The primary outcome was the causal average treatment effect (ATE) for the difference in during-treatment symptom reductions between fCBT and therapist-guided iCBT, with symptoms measured by the Patient Health Questionnaire-9 (PHQ-9). If only one PHQ-9 report was recorded, no change was recorded, thereby penalising dropout. For an optimal and robust (a posteriori-balanced) ATE estimate, we applied targeted maximum likelihood machine learning. There was no involvement of individuals with lived experience in the research and writing process. FINDINGS: The guided iCBT registry recorded patients from Dec 12, 2018, to Dec 22, 2022, and the fCBT registry spanned Aug 28, 2018, to Sept 28, 2022. From the total of 32 343 registered therapies, 392 people were included from the fCBT registry and 5467 people from the iCBT registry. Four people in the fCBT group and 21 in the guided iCBT group had missing baseline data, therefore the main sample for analysis contained 5834 patients (4101 [70%] were female and 1733 [30%] were male) with a mean age of 35 years (SD 12). Altogether, 5455 (94%) patients were registered with Finnish as their native language. The ATE estimate indicated that the PHQ-9 score declined 0·745 points (95% CI 0·156-1·334) more in the iCBT group than in the fCBT group. Sensitivity analyses concurred. INTERPRETATION: Considered alongside previous RCTs, our findings suggest that short first-line treatments with therapist-guided iCBT are at least as effective and efficacious as fCBT. Our findings eliminate error sources and extend the representativeness of the population for this cost-effective treatment. FUNDING: The Research Council of Finland (Academy of Finland). TRANSLATION: For the Finnish translation of the abstract see Supplementary Materials section.