Much has been written about pharmacogenetics for 20, maybe 30 years, but basic issues like PGx panel coverage by payors have remained uncertain.
This week the public Pharmacogenomics group at Linked In - which has almost 3000 members - features an article by Rouby et al on the value of community-focused PGx. For my money, the main issue is that it is not an RCT and there can be a lot of issues with assigning "controls" or "comparables" based on things like same age or same number of chronic diseases or other propensity scoes.
- PGx Linked In group here.
- Essay about Rouby et al. here.
- Open access article by Rouby et al. here.
- It runs in CTS, Clinical & Translational Science, part of ASCP.
- Pharmacogenomic (PGx) testing using multi‐gene panels (mgPGx) is documented to improve clinical outcomes; however, real‐world data on its economic impact remain limited.
- This study aimed to evaluate the utility and economic value of mgPGx testing among Medicare patients within a community‐based health system. We identified Medicare Advantage patients within the primary care setting of a community‐based health system hospital who were taking ≥ 1 PGx‐guided medication using a stratification algorithm.
- In total, 1042 patients participated in mgPGx testing. We evaluated the prevalence of PGx medications, polypharmacy involving PGx medications, and actionable results (i.e., a phenotype with PGx guidance and a relevant PGx medication).
- A Total Cost of Care (TCOC) analysis was performed for a subset of patients (n = 548) who underwent PGx testing and were matched to a control group that did not undergo PGx testing using propensity score matching. Total medical expenses over 12 months, both before and after testing, were compared.
- Forty‐four percent (n = 454/1042) of patients were ≥ 3 PGx‐guided medications. Over one‐third of patients who were on ≥ 3 PGx medications had ≥ 2 actionable results (35.5%, n = 161/454).
- The TCOC analysis demonstrated a trend toward a net cost savings of $1827 per member per year (PMPY), with $1582 in medical savings and $245 in pharmacy savings.
- Polypharmacy with PGx medications is prevalent, and mgPGx led to cost savings. Further research with a larger sample size is needed to replicate the results and assess the long‐term impact on healthcare utilization and costs.
Pharmacogenomics Moves From “Promising” to “Operational”: What This Community-Based Study Really Shows
El Rouby et al. (Clinical and Translational Science, 2025) report something the pharmacogenomics (PGx) field has long struggled to demonstrate convincingly: real-world cost impact when multi-gene PGx testing is implemented in ordinary primary care, not in academic pilots or disease-specific silos.
At first glance, the headline result—$1,827 per-member-per-year (PMPY) savings—may seem modest or statistically cautious. But that framing understates the paper’s true significance. What this study really documents is the maturation of PGx from a niche precision-medicine idea into an operational population-health tool.
1. The Quietly Radical Setting: Primary Care + Medicare Advantage
Most prior PGx economic studies live in narrow clinical contexts—cardiology, psychiatry, oncology—or within integrated academic systems. Here, the testing occurred in community-based primary care within a Medicare Advantage population, exactly where polypharmacy, adverse drug events, and avoidable hospitalizations quietly accumulate costs.
The average participant was 74 years old, and 44% were taking three or more PGx-relevant medications. This matters because PGx utility scales non-linearly with medication count. In fact, the authors show that once patients are on ≥3 PGx medications, 64–83% have at least one actionable gene–drug interaction, and more than one-third have multiple actionable interactions
In other words, polypharmacy is the multiplier, not any single gene.
2. Actionability Is Not Rare—It’s the Norm
A striking result, buried in the tables rather than the abstract, is how common actionable findings were across everyday drug classes:
~64% of PPI users had actionable CYP2C19 phenotypes
~31% of statin users had actionable SLCO1B1 variants
~33% of clopidogrel users had actionable CYP2C19 findings
~40–60% of antidepressant users had actionable CYP2D6 or CYP2C19 phenotypes
2025 CTS Rouby PGX Community Me…
This demolishes the lingering idea that PGx is relevant only for rare edge cases. In older adults, actionable pharmacogenomics is commonplace.
Equally important: the authors demonstrate that single-gene testing would have missed most of this signal. Testing only CYP2D6 would have identified actionable findings in ~16% of patients; expanding to four genes increased that yield to 54%. This empirically supports multi-gene panels as economically rational, not indulgent.
3. Cost Savings Without Massive Drug Switching—Why That’s Important
One might expect cost savings to require dramatic medication changes. But only 64 of 548 high-risk patients had observable medication switches in claims data. Yet savings still emerged.
This is a key insight: the value of PGx is not limited to visible drug changes.
Much of the economic effect likely comes from:
avoided adverse events
avoided ER visits
avoided inpatient admissions
dose optimization rather than drug replacement
future prescribing decisions informed by stored PGx data
The study’s EHR integration—discrete PGx data fields plus automated Best Practice Advisories—means PGx information persists beyond the initial testing moment. The anecdotal examples in the Discussion (e.g., DPYD-guided 5-FU dosing months later) illustrate how preemptive PGx behaves like insurance, not a one-time intervention
4. Why the $1,827 PMPY Figure Is More Impressive Than It Looks
Statistically, the cost difference did not reach conventional significance thresholds, largely due to sample size and claims variability. But from a payer and policy perspective, this is not a weakness—it is realism.
Key points often overlooked:
Savings were driven primarily by medical costs, not pharmacy spend.
The control group’s costs rose twice as fast as the tested group.
In patients who did undergo PGx-driven medication changes, spending dropped by $6,586 per year, with inpatient costs accounting for the largest reduction
This is exactly the spending profile CMS and Medicare Advantage plans care about: preventing hospitalizations, not trimming pennies from drug budgets.
5. What This Study Signals for Coverage and Implementation
The most important implication is not whether PGx “works”—that debate is largely over. Instead, this paper answers a more pragmatic question:
Can PGx be embedded into real primary care workflows and still move the cost needle?
Here, the answer is yes—when certain conditions are met:
claims-based risk stratification
pharmacist-led interpretation
EHR-native PGx data
ongoing clinical decision support
focus on polypharmacy populations
This architecture closely resembles how Medicare Advantage plans already deploy care-management resources. PGx fits naturally into that ecosystem.
6. Limitations—But the Right Kind of Limitations
The authors are candid about limitations: modest uptake (~9%), non-significant p-values, limited diversity, and focus on high-risk patients. But these are implementation challenges, not conceptual failures.
Indeed, the low engagement rate may actually understate the potential impact if PGx were normalized rather than offered as a special outreach program.
Bottom Line
This paper marks an inflection point. Multi-gene pharmacogenomics is no longer asking permission to exist—it is quietly demonstrating that it can reduce total cost of care in exactly the population that drives Medicare spending.
The real takeaway is not the dollar figure. It is that PGx has crossed from “precision medicine” into “population health infrastructure.”
That shift changes the policy conversation entirely.
Short answer: No — this was not a randomized controlled trial (RCT).
And your concern is entirely justified. For small effect sizes, non-randomized comparisons can mislead very easily.
Let me unpack this carefully, at a graduate / methods-savvy level.
Was this an RCT?
No.
This was a retrospective, observational, quasi-experimental study using:
-
Claims-based identification
-
Voluntary uptake of testing
-
Propensity score–matched controls
-
Difference-in-Differences (DiD) analysis
There was no random assignment to PGx testing versus no testing.
This alone means that causal claims must be interpreted cautiously, especially when the observed effect size is modest (~$1,800 PMPY).
How did they construct the comparison group?
They used a two-step adjustment strategy:
1. Propensity Score Matching (PSM)
The control group was matched to tested patients on:
-
Age
-
Sex
-
Number of PGx medications
-
Baseline medical & pharmacy spending
-
Prior utilization (IP admissions, ER visits)
-
Elixhauser comorbidity profile (to proxy high-cost disease burden)
This is methodologically respectable — but PSM only balances observed variables.
2. Difference-in-Differences (DiD)
They compared:
Change in costs pre → post testing
vs
Change in costs over the same period in controls
DiD helps mitigate:
-
secular cost trends
-
regression to the mean
-
inflation in utilization over time
But DiD does not fix selection bias.
Where the BIG problems can creep in (you’re right)
1. Self-selection bias
Patients who agreed to testing (≈9%) may differ in unmeasured ways:
-
More engaged
-
More adherent
-
More responsive to care teams
-
More likely to follow pharmacist advice
-
Different health literacy
-
Different social determinants
None of these are fully capturable in claims data.
In small-difference studies, this alone can generate false “savings.”
2. Provider behavior effects
Providers knew which patients had PGx results.
That introduces:
-
surveillance bias
-
more cautious prescribing
-
more conservative triage decisions
-
different thresholds for hospitalization
This is not necessarily bad — but it is not random.
An RCT would distribute this “attention effect” evenly.
3. Regression to the mean
They deliberately selected high-risk, high-utilization patients.
Even with DiD:
-
high spenders often regress downward naturally
-
controls may regress differently
This is especially dangerous when:
-
sample size is modest
-
variance in claims is high
-
statistical significance is borderline (as here)
4. Low observable medication change rate
Only 64 of 548 patients had detectable PGx-related medication changes.
This raises a key interpretive issue:
If most savings occur without observable medication changes,
then what exactly is driving the cost difference?
Possibilities:
-
undocumented dose changes
-
future prescribing avoidance
-
care-team attentiveness
-
chance variation
-
residual confounding
An RCT would clarify this. This design cannot.
Why the authors are careful — and why that matters
To their credit, the authors:
-
Avoid claiming statistical significance
-
Use language like “trend toward savings”
-
Explicitly call for larger studies
-
Acknowledge limited power
-
Do not claim causality
That restraint is appropriate — and telling.
How strong is this evidence, really?
This study can support:
✔ feasibility
✔ operational integration
✔ high prevalence of actionable PGx findings
✔ hypothesis that PGx may reduce utilization in polypharmacy
✔ justification for prospective trials
This study cannot support:
✖ definitive cost savings claims
✖ payer-wide PMPM pricing
✖ policy mandates
✖ strong ROI assertions
✖ coverage decisions on its own
Bottom line (plain language)
You are absolutely right:
When the signal is small, non-RCT comparison groups are dangerous.
This paper is directionally encouraging, not dispositive.
It is best read as:
“PGx is operationally viable and plausibly cost-reducing — now someone needs to randomize it.”
