Stepped care for youths at clinical high risk for psychosis: a real-world study

This real-world study demonstrates that a risk calculator-informed stepped care model implemented at the Yale PRIME Clinic is feasible and associated with significant, sustained improvements in a broad range of symptoms for young people at clinical high risk for psychosis.

The Gist

Imagine you are a gardener tending to a special greenhouse filled with young plants that show early signs of a rare, serious disease. You know these plants might get sick, but you don't know which ones will, or how bad it will get.

This study is about a new way of gardening called "Stepped Care," tested at the Yale PRIME Clinic. Instead of giving every plant the same heavy dose of fertilizer (or medicine) right away, the gardeners used a smart weather forecast to decide exactly how much care each plant needed.

Here is the story of how it worked, broken down simply:

1. The Problem: The "One-Size-Fits-All" Trap

In the past, doctors often had to guess. Should they give a young person with early warning signs of psychosis (a break from reality) strong medication immediately? Or just talk therapy?

• Too little help: The person might get worse.
• Too much help: The person might get unnecessary side effects from heavy medication.

It was like trying to fix a leaky roof by either ignoring it or replacing the whole house. We needed a middle ground.

2. The Solution: The "Risk Calculator" GPS

The Yale team built a GPS for mental health. They used a special tool called the NAPLS Risk Calculator. Think of this calculator as a weather app that predicts the chance of a storm (psychosis) hitting a specific person over the next two years.

Based on this "forecast," they created a three-step ladder of care:

• Step 1 (The Light Raincoat): Everyone starts here. It's a short course of education and support. The doctors teach the young person and their family how to understand the symptoms, manage stress, and spot warning signs. It's like putting on a raincoat before the storm hits.
• Step 2 (The Umbrella & Shield): If the "weather forecast" says the risk is low (less than a 33% chance of a storm), the person moves to Cognitive Behavioral Therapy (CBT). This is like an umbrella that helps them navigate tricky thoughts and feelings. It's more hands-on but still non-medical.
• Step 3 (The Storm Shelter): If the forecast says the risk is high (more than a 33% chance), or if the symptoms are very heavy, they move straight to medication (antipsychotics). This is the heavy-duty shelter for when the storm is already raging.

The Magic: The system isn't static. Every six months, they check the "weather forecast" again. If the risk goes down, the person can step down to lighter care. If it goes up, they step up. It's a flexible, moving target.

3. The Experiment: What Happened?

The researchers watched 71 young people (ages 12–25) for two years. They wanted to see if this "smart ladder" worked.

• The Result: It worked beautifully.
  • Symptoms faded: The "storm clouds" (hallucinations, confusing thoughts, depression) cleared up significantly for most people.
  • Speed: By 6 months, the positive symptoms (the weird thoughts) started getting better. By 2 years, the improvement was strong and lasting.
  • The "Dropouts": About 65% of people didn't stay for the full two years. Usually, when people quit therapy, we worry they quit because they got worse. But here, the data showed something surprising: The people who left had similar improvement trajectories to those who stayed.
  • Why did they leave? It seems many left because they were doing too well. They felt better, got back to school or work, and didn't feel they needed the clinic anymore. It was a "happy dropout."

4. The Takeaway

This study proves that you don't need to blast everyone with heavy medication immediately.

• Precision Medicine: By using a calculator to predict risk, doctors can give the right amount of help at the right time.
• Safety: It avoids over-treating people who might get better on their own or with just a little support.
• Hope: It shows that with the right, tailored support, young people at risk can recover and thrive.

In a nutshell: This study showed that treating young people at risk for psychosis is like tuning a radio. Instead of blasting static (heavy meds) or having no signal (no help), this "Stepped Care" model lets you find the perfect frequency for each person, turning the volume up or down as the signal changes. And the result? A much clearer, happier signal for everyone.

Technical Summary

1. Problem Statement

Young people at Clinical High Risk (CHR) for psychosis experience significant distress and functional decline, with a substantial risk of converting to a full psychotic disorder. While experts recommend Stepped Care (SC) models—tiered systems that calibrate treatment intensity to individual need—few programs have published detailed clinical models or real-world outcomes. Existing interventions lack consistent evidence of durable effects, and there is a critical need to evaluate whether risk-stratified SC models, particularly those informed by prognostic tools like the NAPLS Risk Calculator, are feasible and effective in specialized outpatient settings.

2. Methodology

Study Design & Participants

• Design: A prospective, uncontrolled, real-world observational study conducted over two years.
• Setting: Yale PRIME Clinic (a specialized outpatient clinic in New Haven, CT).
• Sample: 71 individuals (ages 12–25) meeting CHR criteria based on the Structured Interview for Psychosis-Risk Syndromes (SIPS).
• Inclusion Criteria: Ages 12–25, IQ ≥70, English fluency, informed consent capability, and active CHR symptoms.
• Duration: Participants were assessed at baseline and at 2, 6, 12, 18, and 24 months.

Intervention: The PRIME Stepped Care Model
The model utilizes the NAPLS Risk Calculator to determine treatment intensity. All participants began with Step 1, and stepping decisions were made based on risk scores and clinical judgment at 6-month intervals.

• Step 1 (Brief Supportive Psychoeducation - BSP): A flexible, 3-session psychoeducational intervention covering the vulnerability-stress model, coping strategies, and risk factors.
• Step 2 (Personalized CBT for Early Psychosis - P-CBT-EP): A manualized cognitive-behavioral intervention (approx. 15 sessions) targeting cognitive distortions and positive symptoms.
  • Stepping Logic: Participants with a risk score <33% after Step 1 received Step 2 alone. Those with ≥33% risk or high distress received Step 2 + Step 3.
• Step 3 (Antipsychotic Medication): Reserved for high-risk individuals (risk score ≥15% at subsequent assessments) or those with severe symptoms. Medication choice was based on client preference and minimal effective dosing.
• Adjunctive Care: Available across all steps (pharmacotherapy for comorbidities, peer/family support, vocational support).

Statistical Analysis

• Primary Analysis: Linear mixed-effects models (LMM) were used to examine changes in symptom severity (SIPS scales and PHQ-9) over time, adjusted for age and sex, under an Intent-to-Treat (ITT) framework with Missing-At-Random (MAR) assumptions.
• Sensitivity Analyses:
  • Pattern Mixture Models: To test if clinical trajectories differed between treatment completers and non-completers (Missing-Not-At-Random assumption).
  • Brown-Forsythe Test: To evaluate heterogeneity of variance (variability in outcomes) between completers and non-completers.

3. Key Contributions

• Implementation of Risk-Calculator-Guided SC: This is one of the first real-world studies to operationalize a stepped care model where specific risk thresholds (e.g., 33% and 15% conversion risk) dictate the escalation or de-escalation of treatment intensity.
• Broad Symptom Scope: Unlike many studies focusing solely on positive symptoms, this study evaluates improvements across positive, negative, disorganized, general, and depressive symptom domains.
• Retention Analysis: The study specifically investigates the clinical trajectories of those who disengaged from treatment, challenging the assumption that early dropout is solely driven by treatment failure or symptom worsening.

4. Results

Demographics & Retention

• Sample: 58% male; average age 15.6 years.
• Completion: Only 35.2% (25/71) completed the full 24-month protocol.
• Disengagement: 16.9% discharged due to full remission; 26.8% voluntarily withdrew; 14.1% discharged for no-shows; 7% converted to psychosis.
• Overall Retention: 56% remained in care for at least one year.

Clinical Outcomes (Hypothesis 1)

• Significant Improvement: Participants demonstrated significant and sustained reductions in all symptom domains over 24 months.
  • Positive Symptoms: Significant reduction by 6 months, continuing through 24 months.
  • Depressive Symptoms: Significant improvement by 2 months.
  • Negative/General/Disorganized: Significant improvements observed by 12 months.
• Remission Rates: By 24 months, 50% of completers achieved full remission (defined as 6 consecutive months of low positive symptoms).
• Conversion Rate: The observed conversion rate was 7.04%, significantly lower than the baseline calculated risk of ~25% (p < .001).

Retention and Trajectory Analysis (Hypotheses 2 & 3)

• Trajectory Similarity: Pattern mixture analyses revealed no significant difference in average symptom trajectories between completers and non-completers. Non-completers did not show signs of clinical worsening prior to dropout.
• Variability: Non-completers exhibited significantly higher heterogeneity (variability) in their positive symptom and depressive (PHQ-9) trajectories compared to completers.
• Negative Symptoms: Non-completers actually showed lower negative symptoms at 6 and 12 months, suggesting disengagement may be driven by functional improvement (e.g., returning to school/work) rather than treatment failure.

5. Significance and Limitations

Significance

• Feasibility: The study demonstrates that a risk-calculator-informed stepped care model is feasible in a specialized outpatient setting.
• Efficacy Signal: The model is associated with broad symptom relief and a conversion rate significantly lower than predicted naturalistic risk, suggesting the intervention may be effective.
• De-stigmatizing Dropout: The finding that non-completers had similar or better trajectories than completers suggests that early disengagement is not necessarily a sign of treatment failure, potentially guiding more flexible retention strategies.

Limitations

• Lack of Control Group: As an uncontrolled pre-post study, improvements could be attributed to regression to the mean, natural remission, or non-specific therapeutic factors.
• Assessment Bias: Assessments were conducted by the treating psychotherapists, introducing potential allegiance or expectancy effects.
• Generalizability: The model was implemented in a highly specialized clinic with extensive training; feasibility in community settings with fewer resources is unknown.
• Data Gaps: Systematic data on intervention fidelity, patient satisfaction, and the specific contributions of individual steps were not collected.

Conclusion
The study provides preliminary evidence that integrating individualized risk prediction with stepped care is a viable strategy for treating young people at CHR, resulting in sustained symptom reduction and low conversion rates. Future randomized controlled trials with blinded raters are required to confirm efficacy and isolate active treatment components.