63/100

Moderate evidence

Cite
More
EmailLinkedInX

Study of 1,010 therapy patients finds a session-by-session score can flag who may not respond

Study title: The Session-specific Risk Index (SRI) in psychotherapy: Developing longitudinal session-by-session machine learning algorithms to predict the likelihood of non-response to treatment

Evidence type
Observational study
Sample
n = 1,010
Sources
1 sourcePeer-reviewed
Published
2 min read

Study title: The Session-specific Risk Index (SRI) in psychotherapy: Developing longitudinal session-by-session machine learning algorithms to predict the likelihood of non-response to treatment

Researchers built the Session-specific Risk Index (SRI), a score that gives the updated probability that a psychotherapy patient will not respond to treatment. It combines baseline information with clinical data collected in each session. The work used observational records from an outpatient clinic covering 1,010 patients.

Read the study abstract

Introduction This study aimed to develop and introduce an implementation-oriented, session-specific metric called the Session-specific Risk Index (SRI), which defines the likelihood of treatment non-response in psychotherapy. The SRI integrates baseline information with real-time clinical data from each session, providing a continuously updated probability of non-response to treatment. Methods We used observational data from an outpatient clinic, including 1,010 psychotherapy patients. We compared the performanc…

The gain

In an observational study of 1,010 outpatient psychotherapy patients, an elastic net model predicted non-response by session 20 with accuracy of .65-.70 and AUC of .70-.77 across the first ten sessions.

In brief

The team compared elastic net regressions with gradient boosting machines for predicting non-response by session 20. Elastic net models without interaction effects did best over the first ten sessions, with accuracy of .65-.70 and AUC rising from .70 to .77.

The authors say that if clinics adopt the index, it might improve precision in mental health care beyond models that personalise treatment only from baseline information. The index has not yet been tested in routine clinical practice.

Main points

  1. The data came from an outpatient clinic and included 1,010 psychotherapy patients.
  2. Models used initial sociodemographic, diagnostic and clinical information plus the evolution of symptoms across sessions.
  3. Elastic net regressions and gradient boosting machines were compared for predicting non-response by session 20.
  4. Elastic net models without interaction effects performed best over the first ten sessions, with accuracy of .65-.70.
  5. AUC ranged from .70 (95% CI .70-.71) to .77 (.76-.78) over those sessions.
  6. Maximum Youden indices set thresholds that balance sensitivity and specificity to flag high-risk patients.
  7. Midpoints of SRI groups correlated with the share of non-responders in each group (weighted correlation .85, 95% CI .83-.86).

The full story

The SRI is designed as an implementation-oriented metric. Instead of a single baseline prediction, it gives a continuously updated probability of non-response as each session adds clinical data.

To build it, the researchers trained elastic net regressions and gradient boosting machines on records from 1,010 outpatients. Inputs were initial sociodemographic, diagnostic and clinical information combined with how patients' symptoms changed over time. The outcome was non-response by session 20.

The best-performing algorithms were used to define the index. Optimal risk thresholds were set using maximum Youden indices to identify high-risk patients. Higher SRI groups had a higher proportion of patients who did not respond, with a weighted correlation of .85.

Sources

1 on file
  1. Psychotherapy and Psychosomaticsdoi:10.1159/pps/aemag026
    Peer-reviewed2026-11-17

Talk it over

Three questions to start on, two more if you have time
The Session-specific Risk Index reached AUC of .70 to .77 over the first ten sessions. Is that accuracy enough to change how a therapist treats a patient flagged as high-risk, and what would you want the flag to trigger?

Start with a question

Study of 1,010 therapy patients finds a session-by-session score can flag who may not respondhttps://truvace.com/item/the-session-specific-risk-index-sri-in-psychotherapy-developing-longitudinal-ses

  1. The Session-specific Risk Index reached AUC of .70 to .77 over the first ten sessions. Is that accuracy enough to change how a therapist treats a patient flagged as high-risk, and what would you want the flag to trigger?

    Answer in the thread
  2. The data came from one outpatient clinic with 1,010 patients, and the index has not been tested in routine practice. How might results differ for patients or clinics unlike that one?

    Answer in the thread
  3. If you were a patient flagged by the index as likely not to respond by session 20, how would you want to be told, and what would you want to happen next?

    Answer in the thread
If you have more time
  • The researchers compared elastic net regressions with gradient boosting machines, and elastic net did best. What would you want to look up about why a simpler model might win here?

    Answer in the thread
  • What kind of trial, comparing clinics that use the index with clinics that do not, would show whether the score actually improves outcomes for patients?

    Answer in the thread

What readers say