For many life sciences organizations, therapy switching remains a familiar yet insufficiently explained phenomenon. Dashboards capture that patients are switching. Reports quantify how many. But the more critical questions: why it happens, when it happens, and what could have been done differently often remain unanswered.
Consider a typical patient journey.
And yet, for most organizations, these signals remain hidden in plain sight.
Without a unified view of patient-level journeys, stakeholders are left interpreting switching behavior through fragmented lenses—clinical data in one place, access signals in another, and prescribing patterns elsewhere. The result is a reactive approach to therapy management, where interventions occur after the switch, rather than at the moment of emerging risk.
This gap presents both a challenge and an opportunity.
The Customer
In response to this gap, a leading global top-5 pharmaceutical company in oncology sought to better understand early therapy transitions within its ADT portfolio. While adoption remained strong, unexplained switching patterns signalled a deeper, unmet need.
The objective was to uncover the why behind switching; identifying early risk signals and crafting more proactive, patient-centric interventions to improve treatment continuity and overall patient outcomes.
Challenges
While the objective was clear, understanding therapy switching in prostate cancer remained complex.
Existing analyses were largely limited to surface-level metrics such as discontinuation rates and overall adherence, without clarity on why, when, or under what circumstances patients switch. Additionally, clinical, pharmacy, and payer data existed in silos, making it difficult to view the complete patient journey.
There was also an over reliance on descriptive analysis, which often masked early warning signals and patient-level variation. Factors such as payer restrictions, prior authorization, treatment gaps, HCP behavior, and patient-related patterns were not evaluated together.
Dealing with these multiple data points through traditional approaches made it difficult to generate clear and actionable insights.
Our Approach
To move beyond surface-level metrics and understand why therapy switching occurs, we designed an approach combining patient journey analysis with machine learning (ML) using real-world data. Our focus was to evaluate switching behavior at a patient level and identify the most influential drivers.
Patient Level Data
Data Cleansing
- Descriptive Analysis
- Statistical Analysis
Model Development
Model Evaluation
- Key Drivers Identification
- Switch Segments Prediction
- Switch Driver Simulation
- AI Intelligence
Integrated medical and non-medical signals from real-world data
We combined medical and non-medical evidence from real-world data, including payer channel, treatment gaps, duration on therapy, disease stage progression, specialty and account type, along with other available patient-level data points.
Defined patient cohorts and switch event
We classified patients into switch-out and non-switch groups based on therapy continuation. For patients with multiple transitions, we focused only on the first switch event to maintain clarity around the initial therapy change.
Conducted patient journey analysis
We moved beyond aggregate snapshots and performed patient-level treatment journey analysis, examining when switching occurred relative to treatment initiation, disease progression, payer channel, treatment gaps, duration on therapy, and treatment continuity.
Applied descriptive, statistical, and machine learning methods
We conducted descriptive and statistical analyses during data preparation and developed multiple machine learning models to evaluate switching behavior.These models enabled the transition from descriptive insights to predictive intelligence, identifying patterns that are not visible through traditional analytics.
Identified key drivers using Random Forest
We used the selected Random Forest model to derive feature importance scores, identifying the most influential drivers of therapy switching and analyzing patterns such as early vs late switching, planned vs unplanned transitions, and switching within 60 days.
Developed patient-level switch risk scoring and segmentation
Used machine learning–derived probabilities to quantify switching likelihood based on clinical, behavioral, access, and physician-driven factors. Predicted probabilities were normalized on a 0–100% scale to create switch segments.
Introduced switch driver simulation for intervention planning
A simulation layer was introduced to evaluate how switching risk changed under different intervention scenarios. By modifying key drivers such as treatment gaps, physician engagement, and access constraints, the model simulated the impact of potential actions on therapy switching behavior for proactive strategy planning.
Enabled GenAI-powered intelligence
Finally, we integrated a GenAI-powered interface that allowed stakeholders to query switching patterns using natural language. The interface summarized key trends, interpreted risk scores, identified influential drivers, and generated actionable recommendations from complex datasets.
Outcomes
The analysis demonstrated that switch-out behavior is driven by a combination of patient experience, clinical decision-making, and system-level factors.
- Treatment Gap Before Switch52%
- Payer Channel12%
- Treatment Duration Before Switch9%
- Disease Stage Progression7%
- Rejection6%
- Specialty6%
- Account Type4%
- Total Pt Oop Amount2%
- Patient Age Group0.5%
- Patient Comorbidities0.4%
Commercial
Payer Channel
<60 days
Treatment Gap
Urologists/ Urology Centers
Specialty/Account Types
Plan Limitations
Claim Rejection
No Progression
Disease Advancement
Patients often transitioned therapies without treatment gaps, indicating planned decisions influenced by dissatisfaction, medical or non-medical reasons. Payer-related factors such as coverage limitations and prior authorization played a key role, particularly among commercial and Medicare patients. Additionally, differences in prescribing patterns across urology-led settings, along with factors such as comorbidities and age-based trends, further shaped therapy transitions.
53% of patients under commercial payers switched therapies, compared to 40% under Medicare, reflecting the influence of coverage limitations and prior authorization
- Commercial53%
- Medicare40%
- Medicaid4%
- Others1%
- Unmapped1%
- Plan Limitations Exceeded119
- Product/Service Not Covered76
- Prior Authorization Required60
- Refill Too Soon20
- Non-Matched Pharmacy Number16
Within the first 60 days of treatment, switching emerged as a critical pattern, influenced by tolerability, unmet expectations, or adverse events.
- <60 days52%
- 60-90 days10%
- >90 days38%
Disease stage progression, although observed in a smaller cohort, remained a significant driver of switching
- Non-Meta>>>Meta42%
- No Disease Progression58%
Urologists from (hospitals and clinics) were associated with higher switch-out patterns, prescribing of alternative ADTs
- Urologists46%
- Hematologists25%
- Other Specialty17%
- Medical Oncologists8%
- Radio Oncologists4%
- Urology Community48%
- Oncology Community24%
- Academic17%
- Others11%