An Assessment of the Real-World Data Platform TriNetX for Measuring the Association Between Group A Streptococcus and Neuropsychiatric Diagnoses

This retrospective cohort study utilizing the TriNetX real-world data platform found that a positive Group A Streptococcus test was associated with a modestly increased risk of incident ADHD but failed to detect most established poststreptococcal autoimmune conditions, highlighting both the potential and limitations of large health care databases in evaluating postinfectious neuropsychiatric risks.

The Gist

Imagine the human body as a complex city, and the immune system as the city's police force. Sometimes, a common criminal like Group A Streptococcus (GAS)—the bacteria behind strep throat—gets caught. Usually, the police handle it, and the city goes back to normal. But sometimes, the police get confused and start attacking the city's own buildings, causing damage to the heart, kidneys, skin, or even the brain. This is what scientists call an "autoimmune reaction."

For years, researchers have wondered: Does this bacterial confusion ever cause the brain's "control center" to glitch, leading to behavioral issues like ADHD, OCD, or anxiety?

To find out, a team of researchers used a massive digital library called TriNetX. Think of TriNetX as a giant, real-time map of medical records from hundreds of hospitals across the US. It contains data on millions of patients, allowing scientists to look for patterns without having to interview every single person.

The Experiment: A Detective's Test

The researchers didn't just look for GAS and brain issues; they set up a "test drive" to see if the TriNetX map was accurate enough to find the truth. They used three types of "landmarks" to check the map's reliability:

1. The "Known Truths" (Positive Controls): These are connections everyone already agrees on. For example, GAS is known to cause a skin condition called guttate psoriasis and a specific heart/joint condition called rheumatic fever. If the map couldn't find these, the map would be useless.
2. The "False Alarms" (Negative Controls): These are things that definitely shouldn't be linked to GAS, like breaking an arm or having a harmless skin mole. If the map said GAS causes broken arms, the map would be broken.
3. The "Mystery" (The Real Question): Does GAS cause neuropsychiatric issues like ADHD, OCD, or anxiety?

They also ran a parallel test using a different germ, the Epstein-Barr Virus (EBV), which is known to be linked to diseases like Multiple Sclerosis (MS) and Lupus. This was like checking if the map could find a different set of known landmarks to ensure the tool itself was working correctly.

What They Found: The Map's Strengths and Blind Spots

1. The Map Worked for Common Things
When they looked for the "Known Truths," the map found a link between GAS and the skin condition guttate psoriasis. It also found a small but real link between GAS and ADHD. This suggests the map is good at spotting common conditions that happen frequently in primary care settings.

2. The Map Missed the Rare and Complex
Here is where the map hit a wall. Even though science says GAS should be linked to severe conditions like rheumatic fever or Sydenham chorea (a movement disorder), the map did not find them.

• Why? The researchers explain that these conditions are rare, complex, and often diagnosed in specialized clinics that aren't always part of the TriNetX network. It's like trying to find a specific, rare type of tree in a forest using a map that only highlights the most common pine trees. The data simply wasn't detailed enough or the sample size too small to catch these rare events.

3. The Map Didn't Make Up Fake Links
When they checked the "False Alarms," the map correctly said, "No, GAS does not cause broken arms." This is good news; it means the tool isn't just randomly guessing connections.

4. The EBV Test Was Surprising
When they tested the EBV virus, the map failed to find the known link to Lupus or MS. In fact, it oddly suggested EBV might lower the risk of MS. The researchers noted this was unexpected and likely due to the same limitations: the data might not be capturing these complex, long-term autoimmune stories correctly.

The Bottom Line

The researchers concluded that while TriNetX is a powerful tool for spotting common links (like GAS and ADHD or GAS and a skin rash), it currently struggles to detect rare, complex, or long-delayed autoimmune reactions involving the brain.

Think of it like a high-resolution camera: it takes great pictures of big, bright objects (common diseases), but if you try to take a picture of a tiny, faint firefly in the dark (rare autoimmune brain conditions), the camera might miss it entirely. The study doesn't prove or disprove that GAS causes brain issues; rather, it proves that this specific digital map isn't the right tool to find those specific answers yet.

Important Note: The paper explicitly states this is a preprint that hasn't been peer-reviewed yet and should not be used to guide clinical practice. It is a study about the tool (the database), not a final medical verdict on how to treat patients.

Technical Summary

1. Problem Statement

While epidemiological studies and animal models suggest a link between Group A Streptococcus (GAS) infections and neuropsychiatric disorders (such as PANDAS/PANS, OCD, tics, and ADHD), establishing causality in clinical populations remains difficult. Challenges include the high prevalence of GAS infections, the rarity of autoimmune sequelae, and the latency between infection and symptom onset.

Furthermore, there is a lack of large-scale real-world data (RWD) analyses within the US healthcare system to validate these associations. The authors sought to evaluate the utility of TriNetX, a federated network of US healthcare organizations, to:

1. Assess the association between positive GAS tests and incident neuropsychiatric diagnoses within one year.
2. Validate the platform's ability to detect "ground truth" associations (both positive and negative) to determine its reliability for studying complex post-infectious autoimmune conditions.

2. Methodology

Study Design & Data Source

• Type: Retrospective cohort study.
• Database: TriNetX US Collaborative Network, comprising data from 70 healthcare organizations (HCOs), primarily academic medical centers.
• Population: Patients aged 5–18 (for GAS cohort) and all ages (for EBV cohort) with documented laboratory tests.
• Inclusion Criteria: Patients must have at least one "well-visit" (Z00 code) to approximate a primary care population and exclude those with prior psychiatric diagnoses or pervasive developmental disorders.
• Exposure: Positive vs. negative laboratory tests for GAS and Epstein-Barr Virus (EBV).

Cohorts

• GAS Cohort: 178,301 positive tests vs. 411,053 negative tests (pre-matching).
• EBV Cohort: 70,217 positive tests vs. 86,909 negative tests (pre-matching).
• Matching: Propensity score matching (1:1) using a logit model with covariates (age, sex, race, ethnicity) to eliminate bias. Post-matching sizes were 178,301 (GAS) and 64,854 (EBV).

Outcome Measures & Validation Strategy
The study utilized a "Ground Truth" framework to validate TriNetX's sensitivity and specificity:

1. Positive Controls (Expected Associations):
   • GAS: Acute Rheumatic Fever (ARF)/Sydenham Chorea (SC), Poststreptococcal Reactive Arthritis (PSRA), Poststreptococcal Glomerulonephritis (PSGN), Guttate Psoriasis (GP).
   • EBV: Multiple Sclerosis (MS) and Systemic Lupus Erythematosus (SLE).
2. Negative Controls (No Expected Association):
   • Fracture of the humerus shaft and non-neoplastic nevi.
3. Primary Neuropsychiatric Outcomes:
   • Any mental/behavioral/neurodevelopmental disorder, Tic disorders, OCD, Anxiety disorders, Eating disorders, and ADHD.
4. Statistical Analysis:
   • Risk Ratios (RR) and 95% Confidence Intervals (CI) calculated using TriNetX's built-in analytics.
   • Significance threshold: $\alpha = 0.05$.
   • Reliability Check: Analyses were run 5 times over 3 months to account for data fluctuations due to HCO connectivity.

3. Key Results

GAS and Neuropsychiatric Disorders

• Pre-Matching: Showed positive associations for "any mental disorder," tic disorders, OCD, and ADHD.
• Post-Matching: The only statistically significant association remaining was between a positive GAS test and ADHD (RR: 1.09; 95% CI: 1.03–1.15).
• Other Disorders: No significant associations were found for OCD, tics, anxiety, or eating disorders after matching.

Validation of Ground Truths (GAS Cohort)

• Positive Controls:
  • Guttate Psoriasis (GP): Successfully detected (RR: 1.75; 95% CI: 1.06–2.89).
  • ARF/SC, PSRA, PSGN: Failed to detect significant associations post-matching. Case counts were insufficient for PSRA and SC; ARF and PSGN associations disappeared after matching.
• Negative Controls:
  • Non-neoplastic nevus: No association detected (as expected).
  • Humerus fracture: Showed a false positive pre-matching (RR: 1.33), but this was corrected after propensity score matching.

Validation of Ground Truths (EBV Cohort)

• MS: Showed a significant inverse association (decreased risk) post-matching (RR: 0.72; 95% CI: 0.58–0.90), contradicting established literature linking EBV to MS.
• SLE: No association detected.

Data Stability

• Results fluctuated slightly across the 5 iterations due to HCOs being offline for maintenance, though the direction of the ADHD and GP findings remained consistent.

4. Key Contributions

1. Platform Assessment: This is one of the first studies to critically evaluate TriNetX's capability to detect rare, post-infectious autoimmune and neuropsychiatric associations in a US population.
2. Differentiation of Signal vs. Noise: The study demonstrates that while TriNetX can detect common, well-coded conditions (like GP and ADHD), it struggles with rare, heterogeneous, or complex post-infectious conditions (like ARF, PSGN, and MS) likely due to coding limitations and case ascertainment issues.
3. Identification of Limitations: The authors highlight that the 1-year follow-up window may be too short for conditions with delayed diagnosis (like MS or complex neuropsychiatric sequelae) and that the reliance on ICD-10-CM codes in a multi-payer US system introduces significant heterogeneity and potential bias.
4. Refutation of Broad Associations: The study suggests that broad claims linking GAS to "any mental disorder" or specific categories like OCD may not hold up in large-scale US RWD when rigorous matching is applied, with the exception of a modest link to ADHD.

5. Significance and Limitations

Significance

• The findings suggest that TriNetX is a viable tool for studying common post-infectious conditions (e.g., GP) and common neuropsychiatric disorders (e.g., ADHD) but may lack the granularity and coding specificity required for rare or complex autoimmune sequelae.
• The failure to detect established links (e.g., EBV-MS, GAS-ARF) serves as a critical warning for researchers relying solely on administrative databases for rare disease epidemiology without external validation.

Limitations

• Coding Granularity: There are no specific ICD-10 codes for post-streptococcal neuropsychiatric syndromes (PANS/PANDAS), forcing researchers to use broad categories (F01-F99) that may dilute the signal.
• Population Bias: The network is skewed toward academic centers and insured populations, potentially missing community-based cases or those with lower socioeconomic status.
• Follow-up Duration: A 1-year window may be insufficient for conditions that manifest years after infection.
• Data Volatility: The results were sensitive to the real-time availability of participating HCOs, indicating that "real-time" data can be unstable for longitudinal studies.

Conclusion
The study concludes that while TriNetX successfully identified an association between GAS and ADHD and the well-established link to Guttate Psoriasis, it failed to replicate associations for other established post-streptococcal and post-EBV conditions. This indicates that while the platform is useful for common conditions, it currently has limitations in detecting rare, complex, or delayed-onset post-infectious autoimmune disorders in the US healthcare context. Future research requires more specific coding, longer follow-up periods, and potentially prospective designs.