Comparative effectiveness of preferred pharmacological treatment options for bipolar disorder among people with opioid use disorder in British Columbia and Ontario, Canada: protocol for parallel population-based target trial emulations

This protocol outlines a parallel population-based target trial emulation study using linked health administrative data from British Columbia and Ontario (2010–2023) to compare the effectiveness of various pharmacological treatments for bipolar disorder among individuals with concurrent opioid use disorder, with outcomes including psychiatric acute-care visits, treatment discontinuation, and all-cause mortality.

The Gist

Imagine you are trying to fix a car that has two very different, complicated problems at the same time: the engine is misfiring (Bipolar Disorder), and the fuel system is contaminated with the wrong kind of gas (Opioid Use Disorder).

Right now, mechanics (doctors) have a great manual for fixing engines and a great manual for cleaning fuel systems. But they don't have a manual for fixing both at the same time. They are guessing which combination of tools works best, and sometimes the car breaks down even faster.

This research paper is a blueprint for a massive, real-world experiment designed to finally write that missing manual.

Here is the breakdown of what the researchers are planning to do, using simple analogies:

1. The Problem: A "Double Trouble" Diagnosis

People with Bipolar Disorder (BD) often struggle with extreme mood swings (high highs and low lows). People with Opioid Use Disorder (OUD) struggle with addiction. When someone has both, it's like trying to drive a car with a broken engine while the fuel is also on fire.

• The Reality: These patients get worse outcomes, die younger, and are harder to treat.
• The Gap: Doctors don't know which medication works best for this specific "double trouble" group. The usual rules might not apply because the drugs interact with the addiction in weird ways.

2. The Solution: The "Time-Traveling" Experiment

Normally, to find the best medicine, doctors run Randomized Controlled Trials (RCTs). This is like a cooking competition where they force 100 chefs to make the exact same dish with the exact same ingredients to see which tastes best.

• The Problem: You can't force people with complex addictions and mental health issues into a strict cooking competition. They often drop out, or the rules are too strict, so the results don't reflect real life.
• The Workaround: Instead of a cooking competition, these researchers are going to be detectives. They will look at the "receipts" (medical records) of thousands of people who already took different medicines in real life.

3. The Method: Emulating a "Target Trial"

The researchers are using a clever trick called "Target Trial Emulation."

• The Analogy: Imagine you want to know if a specific brand of tire is better for snowy roads. You can't stop time and run a perfect test. Instead, you look at 10,000 cars that already drove through snow. You group the ones with Brand A tires and the ones with Brand B tires.
• The Magic: To make sure the comparison is fair, they use a mathematical "magic lens" (called Inverse Probability Weighting). This lens adjusts the data so that the group with Brand A tires looks exactly like the group with Brand B tires in terms of age, health, and driving habits. This makes it look like a fair experiment, even though it happened naturally.

4. The Two "Test Tracks"

They are running this investigation in two different Canadian provinces: British Columbia (BC) and Ontario.

• Why two? It's like testing a new car on two different tracks: one in the mountains (BC) and one on the flat plains (Ontario). If the car performs well on both tracks, you know it's a truly good car. If it only works on one, you know it's specific to that terrain.
• The Data: They are using massive government databases that track almost every doctor visit, hospital stay, and prescription filled in these provinces.

5. The Four "Races" (Comparisons)

The researchers aren't just testing one drug against another. They are running four separate "races" to see which strategy wins:

1. Race 1: The "Classic Stabilizer" (Lithium) vs. Other Mood Stabilizers (like Valproate).
2. Race 2: The "Classic Stabilizer" (Lithium) vs. Modern Antipsychotics (like Quetiapine).
3. Race 3: The "Classic Stabilizer" (Lithium) vs. Combination treatments (Lithium + Antipsychotic).
4. Race 4: A specific heavy-duty combo (Lithium + Valproate) vs. other heavy-duty combos.

6. What Are They Looking For? (The Finish Line)

They aren't just checking if the patient feels "happy." They are looking at the big, life-or-death metrics:

• The "Emergency Brake": Did the patient have to go to the ER or get hospitalized for a mental health crisis?
• The "Fuel Gauge": Did the patient stop taking the medicine? (Stopping early is a huge problem).
• The "Engine Failure": Did the patient pass away from any cause?

7. The "What-If" Scenarios (Sensitivity Analysis)

Since they are looking at old data, they know things might be messy. So, they are running "What-If" scenarios:

• What if we only look at people who didn't go to jail?
• What if we only look at the years before the pandemic?
• What if we define "stopping medicine" as 7 days instead of 30?
  This ensures that their final answer is solid and not just a fluke of the data.

The Bottom Line

This paper is a plan (a protocol) to solve a mystery that has plagued doctors for years. By using a "detective" approach on real-world data from two huge provinces, they hope to finally tell doctors: "If you have a patient with both Bipolar Disorder and Opioid Addiction, here is the specific medicine that keeps them safe, keeps them on treatment, and keeps them alive."

It's about moving from guessing to knowing, so that the most vulnerable patients get the best care possible.

Technical Summary

Based on the provided protocol paper, here is a detailed technical summary. Note that as this is a study protocol, it outlines the planned methodology and objectives rather than reporting finalized results.

1. Problem Statement

• Clinical Gap: Individuals with concurrent Bipolar Disorder (BD) and Opioid Use Disorder (OUD) face severe clinical outcomes, including higher mortality, complex treatment needs, and delayed diagnosis. However, there is a critical lack of evidence-based clinical guidelines specifically for this population.
• Limitations of Existing Evidence:
  • RCT Exclusion: Randomized Controlled Trials (RCTs), the gold standard, typically exclude individuals with OUD and complex comorbidities, leading to a lack of external validity for this specific group.
  • Observational Weakness: Existing observational studies are often small, uncontrolled, or industry-led, failing to account for the complexity of drug-drug interactions (e.g., between Opioid Agonist Therapy and BD medications) and the masking of symptoms.
  • Guideline Void: Canadian and US guidelines offer no specific recommendations for treating BD in patients with OUD, leaving clinicians to rely on heterogeneous, non-evidence-based approaches.

2. Methodology

The study employs a parallel population-based target trial emulation framework using linked health administrative data from two Canadian provinces: British Columbia (BC) and Ontario.

A. Data Sources & Study Setting

• Provinces: BC (2010–2022) and Ontario (2014–2023).
• Data Linkage: The study links multiple administrative databases, including:
  • Demographics/Registry: Client Roster (BC) / Registered Persons Database (Ontario).
  • Pharmacy: PharmaNet (BC, captures all residents) / Ontario Drug Benefit (ODB) & Narcotics Monitoring System (NMS) (Ontario, captures public coverage and narcotics).
  • Clinical: Medical Services Plan (BC) / OHIP Claims (Ontario), Discharge Abstract Database (Hospitalizations), National Ambulatory Care Reporting System (ED visits).
  • Outcomes: Vital Statistics (Mortality).
  • Unique to Ontario: Ontario Mental Health Reporting System (OMHRS) with RAI-MH clinical scales.
  • Unique to BC: Provincial Corrections data (used as an exclusion criterion in primary analysis).

B. Study Population

• Inclusion: Individuals aged ≥18 with a diagnosis of both BD and OUD who initiate BD pharmacological treatment.
• Exclusion: Schizophrenia, kidney/liver toxicity, cancer/palliative care, pregnancy, and (in primary BC analysis) incarceration.
• Design: Both Incident User (no prior BD treatment history) and Prevalent New User (30-day washout) designs are planned.

C. Target Trial Emulation (Four Parallel Trials)

The study emulates four distinct comparative trials:

1. Lithium vs. Non-antipsychotic mood stabilizers (divalproex, lamotrigine, valproic acid).
2. Lithium vs. Second-generation antipsychotics with mood-stabilizing properties (risperidone, olanzapine, aripiprazole, quetiapine).
3. Lithium vs. Combination therapies (e.g., Lithium + Divalproex, Lithium + Olanzapine, etc.).
4. Lithium + Valproate (LATVAL) vs. Various combination therapies.

D. Analytical Strategy

To address confounding and mimic RCT conditions, the study utilizes a multi-layered causal inference approach:

• Primary Analysis (Initiator/ITT):
  • Inverse Probability Weighting (IPW): Uses investigator-specified covariates (demographics, comorbidities, healthcare utilization) to balance treatment groups.
  • High-Dimensional Propensity Score (hdPS): Incorporates 200 empirical covariates derived from ICD codes and drug dispensations to adjust for unmeasured confounding (e.g., disease severity).
  • Instrumental Variable (IV) Analysis: Uses prescriber and region-level prescribing preferences as instruments to address residual unmeasured confounding via a two-stage residual inclusion (2SRI) approach.
• Secondary Analysis (Per-Protocol):
  • Marginal Structural Models (MSM): Uses inverse probability of censoring weights to handle time-varying confounding caused by treatment switching and dose adjustments.
  • Individuals are censored if they deviate from the assigned strategy (switch meds) or deviate from recommended dosages.

E. Outcomes

• Primary: Time to psychiatric acute-care visits (hospitalization/ED), BD treatment discontinuation (defined as a ≥30-day gap), and all-cause mortality.
• Secondary: Cause-specific mortality (suicide), treatment switching, and clinical symptom scales (available only for Ontario via OMHRS).

3. Key Contributions

• Methodological Rigor: This is one of the first studies to apply a target trial emulation framework specifically to the BD-OUD population, utilizing advanced causal inference techniques (hdPS, IV, MSM) to overcome the limitations of traditional observational studies.
• Parallel Validation: By conducting identical analyses in two distinct jurisdictions (BC and Ontario) with different data structures, the study allows for a direct assessment of the replicability of findings across settings.
• Real-World Evidence: It addresses the "black box" of treating BD in OUD patients by analyzing real-world prescribing patterns and outcomes, filling a critical gap where RCTs cannot reach.
• Granular Stratification: The protocol plans for stratified analyses by BD-I, manic phases, and depressive phases, acknowledging the heterogeneity of the disorder.

4. Results

• Status: No results are reported in this document.
• This is a protocol paper describing the study design, data sources, and analytical plan. The results will be generated upon the completion of the data linkage, cohort creation, and statistical analysis.

5. Significance

• Clinical Impact: The findings will directly inform clinical guidelines for a high-risk, underserved population. If specific treatments (e.g., Lithium vs. Antipsychotics) are found to be superior in reducing mortality or hospitalization, guidelines can be updated to reflect evidence-based care for BD-OUD comorbidity.
• Policy Implications: Results will be disseminated to provincial health ministries and national guideline developers to standardize care across Canada, potentially reducing the 10-year diagnostic delay and improving retention in treatment.
• Public Health: Given the elevated mortality risk (40% higher than those without mental disorders) in this population, identifying effective pharmacological strategies could have a substantial impact on reducing suicide and overdose-related deaths.
• Ethical Framework: The study adheres to the TARGET Statement for reporting observational studies emulating trials, ensuring transparency and reproducibility.

Conclusion: This protocol outlines a robust, large-scale epidemiological study designed to generate high-quality evidence on the comparative effectiveness of BD treatments in patients with OUD, aiming to replace clinical uncertainty with data-driven decision-making.