Paul is 32 and has recently been referred to a community psychosis service. He has been struggling with paranoid thoughts and voices that threaten him. At times he is convinced that people are talking about him or planning to hurt him.
He feels worn down and finds it hard to concentrate. Some days, just leaving the house takes real effort.
At his first appointment, before any therapy begins, he is given several assessment forms. They are long. He fills them in and then waits for them to be reviewed, discussed, and turned into a plan. It’s only after the forms that treatment starts.
Paul is just one example of a common situation in mental health services. Assessment is essential, but when someone is already dealing with distress, voices and low energy, the length and timing of that assessment can add to the burden, instead of supporting recovery.
This raises a practical question: how do we measure symptoms like paranoia accurately and regularly, without increasing the burden on people who are already struggling?
Despite broad agreement that routine outcome measurement matters in psychosis services, putting it into practice has proved difficult. Part of the problem is structural: psychosis is highly varied, with people having very different experiences of paranoia, hallucinations, grandiosity, cognitive disorganisation and other dimensions (Freeman et al., 2021). Consensus based outcome assessments have tended to resolve this by focusing on what’s relevant to everyone. This can mean psychotic experiences themselves get dropped or reduced to a handful of generic items (McKenzie et al., 2022). Meanwhile, comprehensive fixed-format questionnaires covering multiple dimensions can quickly become long and burdensome for patients who are already experiencing distress, and for clinicians trying to use assessment time well.
The result is a gap between what measurement-based care could offer (timely, personalised, treatment-guiding data) and what happens in everyday services. Even when measures are introduced, sustaining their use, and embedding them into everyday clinical decisions, can be challenging (Lewis et al., 2022). Collecting scores does not automatically mean they are discussed with patients or used to guide treatment.
A new study by Freeman and colleagues (2025), published in BMJ Mental Health, explores whether computerised adaptive testing can provide precise estimates of paranoia using just a small number of tailored questions, potentially making routine assessment in clinical settings more feasible.
There is a gap between what measurement based care could offer people experiencing paranoia and what is on offer.
Methods
To examine whether paranoia could be measured more efficiently, Freeman et al. (2025) focused on the 10-item Revised Green et al. Paranoid Thoughts Scale – Part B (R-GPTS; Freeman et al., 2021), a widely used dimensional self-report measure of persecutory thinking. The scale asks participants to rate how strongly they have experienced thoughts such as “Certain individuals have had it in for me” or “I was convinced there was a conspiracy against me” over the past month. Higher scores indicate greater severity of paranoia.
Instead of administering all ten items to every individual, the authors evaluated a computerised adaptive testing (CAT) version. In CAT, each new question is selected based on a person’s previous responses, meaning that only the most informative items are presented.
The adaptive algorithm was built using item response theory (IRT), a statistical framework that estimates how well each item differentiates between levels of severity.
The CAT was evaluated using four existing datasets in which the full R-GPTS had already been administered. These included:
- A large UK adult representative survey (n = 10,382), quota-sampled to match the population on age, gender, ethnicity, income and region;
- 319 adult patients with psychosis taking part in the gameChange clinical trial;
- 836 adult male NHS patients with psychosis attending mental health trusts; and
- 89 patients with current persecutory delusions enrolled in the Feeling Safer clinical trial.
Together, these samples covered the full paranoia continuum, from the general population to individuals experiencing severe delusional beliefs.
CAT simulations were conducted across these datasets. The test ended either when the score was precise enough to be considered reliable, or after five questions.
An adaptive algorithm was built using item response theory (IRT), a statistical framework that estimates how well each item differentiates between levels of severity.
Results
Across all four datasets, the adaptive version performed well.
On average, the CAT administered around four items per person instead of the full ten-item questionnaire, a reduction of more than 50% in assessment length.
Despite this substantial reduction, agreement between the adaptive scores and the full-scale scores remained high:
- r = 0.95 in the general population sample
- r = 0.94 in both psychosis samples
- r = 0.87 in the persecutory delusions sample
In practical terms, this means the shorter adaptive version produced very similar estimates of paranoia severity to the full questionnaire.
Measures of accuracy indicated acceptable levels of error, and systematic bias was minimal. The adaptive test showed a very slight tendency to underestimate paranoia scores, but the difference was small and unlikely to be clinically meaningful.
However, performance was not identical across the entire continuum. Estimates were somewhat less precise:
- Near the boundary between “average” and “elevated” paranoia
- At the highest severity levels
In the representative population sample, approximately 4% of individuals below the “elevated” threshold were classified as elevated by the adaptive test. While this false-positive rate is relatively low, it highlights that dimensional cut-offs should be interpreted cautiously.
Overall, the findings suggest that substantial reductions in assessment length are possible without major loss of psychometric accuracy, at least under simulation conditions.
These findings suggest that substantial reductions in assessment length are possible without major loss of psychometric accuracy, at least under simulation conditions.
Conclusions
Freeman and colleagues conclude that computerised adaptive testing can generate accurate estimates of paranoia across its full severity continuum while substantially reducing assessment length. In both general population and clinical samples, an average of four tailored questions closely approximated scores from the full ten-item scale.
Although precision was slightly lower near certain cut-off points and at the highest severity levels, overall agreement was strong and systematic bias minimal. These findings suggest that adaptive, dimensional assessment of paranoia is technically feasible and may support more practical implementation of routine measurement in clinical settings.
Dimensional assessment of paranoia appears technically feasible.
Strengths and limitations
A key strength of this study is its coverage of the full paranoia continuum. By including both a large representative community sample and multiple clinical groups, including individuals with current persecutory delusions, the authors tested the adaptive approach across a broad and clinically relevant range of severity. The consistency of results across these heterogeneous datasets strengthens confidence in the robustness of the findings.
The psychometric foundation is also solid. The CAT algorithm was built on a well-validated, IRT-calibrated measure (Freeman et al., 2021). For dimensional constructs such as paranoia, IRT is particularly appropriate because it allows items to differ in how well they discriminate across severity levels. In this respect, the statistical method aligns closely with contemporary dimensional models of psychosis.
However, several limitations deserve attention.
First, this was a simulation study. Although simulations are rigorous for evaluating statistical performance, they cannot fully anticipate real-world implementation issues such as patient engagement, digital accessibility, clinician acceptance, or integration within service workflows.
Second, precision was lower near severity thresholds. Small score differences around cut-offs could lead to misclassification. This highlights a broader issue: dimensional scores should inform clinical judgement rather than define it.
Third, while adaptive testing efficiently estimates severity, it does not capture the cognitive, emotional, or social processes that maintain paranoia, such as worry, threat anticipation, anomalous experiences, or safety behaviours. From a clinical psychology perspective, severity scores are informative, but they do not replace an individualised formulation of why the paranoia is occurring and what is maintaining it.
Finally, the study demonstrates psychometric feasibility, but practical feasibility in routine services remains to be tested.
Implications for practice
Persecutory delusions are among the most frequent and distressing psychotic symptoms (Collin et al., 2023). Yet in many psychosis services, outcome monitoring remains broad or infrequent, often relying on global symptom scales rather than assessing specific dimensions.
Focusing on clearly defined symptom dimensions, rather than relying solely on global measures, may be an important first step toward more responsive care. Freeman et al. situate adaptive testing within the broader framework of measurement-based care: the idea that systematic, repeated assessment can guide treatment decisions, monitor progress, and support service-level evaluation.
By reducing the number of items required while maintaining acceptable precision, CAT may lower the burden on patients and clinicians. This is particularly relevant in psychosis services, where heterogeneity is high and comprehensive fixed batteries can quickly become impractical. A brief, adaptive measure of paranoia could realistically be administered:
- At intake
- During psychological therapy
- At review appointments
- Within digital or blended care pathways
Crucially, more efficient measurement may also support more personalised care. If symptom dimensions such as paranoia can be assessed accurately and repeatedly, clinicians may be better positioned to detect early deterioration, identify non-response, and adapt interventions accordingly.
This may be particularly relevant in the context of brief or digitally delivered interventions, including single-session or modular online approaches. When interventions are short and targeted, having a precise, low-burden measure of paranoia could allow clinicians to observe meaningful changes over short timeframes and evaluate whether a specific component is having the intended effect.
However, feasibility is not only technical. Although the infrastructure for adaptive testing already exists, successful implementation would depend on clinician engagement, integration into electronic health systems, and clarity about how scores should inform decisions.
Importantly, severity scores should complement, not replace, collaborative formulation. A rising paranoia score tells us that something has changed; understanding why it has changed, and which mechanisms are involved, remains essential.
Ultimately, the promise of adaptive testing lies not only in efficiency, but in its potential to support more responsive and personalised clinical care. This, and similar research to fine tune and individually adapt assessment, has significant potential to reduce the burden on the people being assessed and on clinicians. This could ensure care is driven by data and responsive to changing symptoms and needs.
Paranoia severity scores should complement, not replace, collaborative formulation.
Statement of interests
Almudena Trucharte conducts research in related areas of paranoia and psychological processes in psychosis. This blog was drafted with the assistance of AI tools for structural support and language refinement; the final content was reviewed, edited, and approved by the author.
Editor
Edited by Simon Bradstreet.
Links
Primary paper
Daniel Freeman, Sinéad Lambe, Felicity Waite, Laina Rosebrock, Anthony Morrison, Kate Chapman, Robert Dudley, Stephanie Common, Julia Jones, Thomas Kabir, Ariane Beckley, Verity Westgate, Natalie Rouse, Bao Sheng Loe (2025) Computerised adaptive testing across the paranoia continuum. BMJ Mental Health, 28, e302099.
Other references
Collin S, Rowse G, Martinez A P & Bentall R P (2023) Delusions and the dilemmas of life: A systematic review and meta-analyses of the global literature on the prevalence of delusional themes in clinical groups. Clinical Psychology Review, 104, 102303.
Freeman D, Loe B S, Kingdon D. et al (2021) The revised Green et al., Paranoid Thoughts Scale (R-GPTS): psychometric properties, severity ranges, and clinical cut-offs. Psychological Medicine, 51, 244–253.
Lewis, C. C., Boyd, M. R., Marti, C. N., & Albright, K. (2022). Mediators of measurement-based care implementation in community mental health settings: results from a mixed-methods evaluation. Implementation Science, 17(1), 71.
McKenzie E, Matkin L, Sousa Fialho L, et al. (2022). Developing an International Standard Set of Patient-Reported Outcome Measures for Psychotic Disorders. Psychiatric Services, 73:249–58.
