How Much Genetic Testing for Medication Costs
Pharmacogenomic testing costs between roughly $50 and $2,000 out of pocket, and the price has almost no relationship to how much information you get back. The widely marketed psychiatric panels cluster around $330 self-pay, with published Medicare reimbursement for one of them near $1,569. Hospital and pharmacy-run panels are often $250 to $400 cash. Direct-to-consumer panels that genotype a handful of positions in five to ten genes sell for under $100. A clinical whole-genome sequence, which contains every position those panels interrogate plus everything else, runs a few hundred to a few thousand dollars depending on depth and who reads it. If you are technically fluent and willing to run open-source tools, the sequencing route is the better buy, with one important exception that we cover below.
What the price tag is paying for
A pharmacogenomic panel is cheap laboratory work wrapped in expensive interpretation. The wet lab step is targeted genotyping: a TaqMan array, a mass-spectrometry genotyping platform, or a small amplicon sequencing panel that reads a fixed list of positions. Reagent and labor cost per sample is low, often tens of dollars at volume.
Everything above that is interpretation and reporting infrastructure, plus sales and prior-authorization overhead. The high-priced psychiatric panels bundle a proprietary drug-by-drug categorization into green, yellow, and red bins across dozens of medications. That categorization is the product. The underlying genotypes are commodity.
This matters for how you shop. If you want the categorized report handed to a prescriber who does not want to read star alleles, you are buying a service and the price is what it is. If you want data, you are overpaying by an order of magnitude.
The genes that carry almost all of the signal
Most of the actionable pharmacogenomics in routine prescribing sits in a short list. The Clinical Pharmacogenetics Implementation Consortium (CPIC) publishes gene-drug guidelines with evidence levels, and the level A set is where the real content lives.
For psychiatric medications, the genes that drive dosing guidance are CYP2D6 and CYP2C19. These two enzymes metabolize a large fraction of antidepressants and antipsychotics. Other specialties have their own recurring names. They include CYP2C9 and VKORC1 (warfarin), SLCO1B1 (statin myopathy risk), and DPYD (fluoropyrimidine toxicity). The list continues with TPMT and NUDT15 (thiopurines), CYP3A5 (tacrolimus), and UGT1A1 (irinotecan). The HLA alleles HLA-B*57:01 and HLA-B*15:02 cover severe hypersensitivity reactions.
A $60 panel and a $1,500 panel usually cover the same core of that list. The expensive one adds more genes with weaker evidence and a longer drug table. Additional genes with no level A guideline add report length. They do not add decisions.
Where cheap genotyping quietly fails
The interesting engineering problem in pharmacogenomics is CYP2D6, and it is where price and quality genuinely diverge. CYP2D6 has over one hundred named star alleles, sits next to two highly similar pseudogenes (CYP2D7 and CYP2D8), and varies in copy number from zero to more than five. Whole-gene deletions, whole-gene duplications, and CYP2D6-CYP2D7 hybrid alleles are common and change the predicted metabolizer phenotype outright.
A fixed genotyping array that reads twelve single-nucleotide positions cannot see copy number and cannot resolve hybrids. The array will report a diplotype that looks confident and is incomplete. Labs that do CYP2D6 well add a copy-number assay, usually quantitative PCR against a reference locus, or use long-read or long-amplicon sequencing. When comparing two panels at different prices, the question to ask is narrow: do you call CYP2D6 copy number and hybrid alleles, and by what method?
The second failure mode is phasing. Metabolizer phenotype depends on which variants sit on the same chromosome. Unphased genotypes force the lab to guess the most probable diplotype. Short-read sequencing has the same limitation unless you add a phasing step or long reads.
The third is ancestry coverage. Panel content was largely defined on European-ancestry variant frequencies. Star alleles common in African, East Asian, and Indigenous populations are underrepresented on older arrays. A no-call in a relevant gene is often reported as the reference allele by default, which biases results toward normal metabolizer status. The same ancestry portability problem shows up across genomic prediction more broadly, and it is the central practical constraint on polygenic scores in clinical use.1
Getting the same calls out of a whole genome
If you have a whole-genome sequence, you already own the raw material. What you need is a caller that understands star allele nomenclature.
Start from a CRAM or BAM aligned to GRCh38 and a joint or single-sample VCF. Thirty-fold mean coverage is enough for most genes. The current standard open-source stack is PharmCAT. It takes a VCF and normalizes it against the CPIC and PharmVar allele definitions. It then calls diplotypes and emits CPIC and DPWG guideline text. The wrapper handles preprocessing:
pharmcat_pipeline sample.vcf.gz \
-o ./pharmcat_out \
-reporterJson
The preprocessor needs a bgzipped, indexed VCF with the reference genome available, and it will fail loudly if your contig names mix chr1 and 1 styles. Fix that upstream with bcftools annotate --rename-chrs rather than editing by hand.
PharmCAT deliberately does not call CYP2D6 from short reads on its own. Run Cyrius separately, which models copy number and hybrid structures from read depth and discriminating positions:
star_caller.py --manifest bams.txt --genome 38 \
--prefix sample --outDir ./cyrius_out
Then feed the Cyrius diplotype back into PharmCAT with -matcherHtml off and the outside-call file option so the final report is internally consistent. Aldy and PyPGx are reasonable alternatives and useful as a second opinion. When two callers disagree on CYP2D6, treat the result as uncertain and stop there.
HLA typing is a separate problem. Short-read HLA callers such as OptiType and HLA-HD work on the class I loci, but for HLA-B*57:01 and HLA-B*15:02 we would confirm with a dedicated clinical assay before anyone acts on it. The clinical consequence of a wrong call is severe enough that a research-grade call is not sufficient.
The case for and against sequencing instead of a panel
The argument for sequencing is that the pharmacogenomic answer is a byproduct. One assay gives you star alleles today plus the ability to re-analyze when CPIC adds a gene-drug pair in three years, without a new blood draw. Genotype is also fixed for life, which is what makes it a stable exposure variable rather than a moving biomarker.2
The argument against is that a genome returns findings you did not ask for. The BabySeq trial found that genome sequencing in apparently healthy infants returned unanticipated monogenic disease risk findings in a meaningful minority of participants, and that returning them required real clinical infrastructure.3 You should decide in advance whether you want that, and you should have a clinician and ideally a genetic counselor available if you do.
There is also the plain fact that a research-grade pipeline you ran yourself is not a clinical result. Laboratories operating under CLIA certification carry validation, quality control, and accountability that your laptop does not. If a diplotype is going to inform a prescription, it should be confirmed in a clinical lab and interpreted by the prescriber. Pharmacogenomic results change dosing decisions, and dosing decisions belong to a clinician.
Questions people also ask
How much does genetic testing for medication cost? Expect $50 to $2,000 out of pocket, with most clinical panels between $250 and $400 self-pay. Insurance coverage varies by payer and by indication, and coverage for psychiatric panels remains inconsistent.
Why is psychiatric pharmacogenomic testing priced higher? The marketed psychiatric panels bundle a proprietary drug categorization and a sales and prior-authorization operation into the price. Published Medicare reimbursement for one such test has been about $1,569, while the same vendor’s self-pay rate is $330.
Are my 23andMe or AncestryDNA raw data enough? They are useful for a rough CYP2C19 or CYP2C9 look and not adequate for CYP2D6, because the arrays do not measure copy number. Array data also has a non-trivial per-genotype error rate at rare positions, so treat any single call as a hypothesis.
Do I need to repeat the test? No. Your germline genotype does not change, so the assay is once per lifetime. What changes is the interpretation, which is why keeping the raw VCF or CRAM matters more than keeping the PDF.
Is it HSA or FSA eligible? Pharmacogenomic panels ordered through a clinician are generally eligible, and several hospital programs state this explicitly. Confirm with your plan administrator before assuming it.
Footnotes
-
John L. Slunecka, Matthijs D. van der Zee, Jeffrey J. Beck, et al. Implementation and implications for polygenic risk scores in healthcare. Human Genomics, 2021. https://doi.org/10.1186/s40246-021-00339-y ↩
-
George Davey Smith, Shah Ebrahim. ‘Mendelian randomization’: can genetic epidemiology contribute to understanding environmental determinants of disease?*. International Journal of Epidemiology, 2003. https://doi.org/10.1093/ije/dyg070 ↩
-
Hadley Stevens Smith, Bethany Zettler, Casie A. Genetti, et al. The BabySeq Project: A clinical trial of genome sequencing in a diverse cohort of infants. The American Journal of Human Genetics, 2024. https://doi.org/10.1016/j.ajhg.2024.08.011 ↩
Pothos builds longitudinal molecular profiles of individuals: whole-genome sequencing, single-cell RNA sequencing, and plasma proteomics, delivered as a detailed report, every data file, and stored cells. Build your profile.