A drug’s addressable market is the patient population that fits its intended use, narrowed by clinical eligibility and the ability to reach treatment. Disease prevalence is only the starting point. A forecast must distinguish people living with a condition from those who are diagnosed, meet the treatment criteria, can obtain care and are likely to start the drug within a defined period. Those denominators produce very different numbers. Multiplying headline prevalence by an annual drug price skips most of the commercial work.
For biotech investors, market sizing connects the clinical evidence to the revenue opportunity. The task is to build a patient model whose assumptions can be checked, challenged and updated. Within biotech commercialisation and market economics, this is the bridge between a promising treatment and a defensible sales forecast. The method below starts with epidemiology, moves through eligibility and access, then converts treatment activity into revenue. All numerical examples are hypothetical, not estimates for an actual medicine.
Define the market before counting patients
Write the proposed market definition before opening a spreadsheet. It should identify the country, forecast year, disease subtype, age group, disease stage, treatment setting and any required biomarker. Also state whether the model counts people eligible at a particular date, patients becoming eligible during a year, treatment courses or patients receiving ongoing therapy. These are different units. “Adults with disease X in the United States” is too broad for a drug intended for biomarker-positive patients whose disease has progressed after an earlier therapy. The narrower definition may produce a smaller number, but it gives investors something useful to value.
| Population | What it measures | What it does not establish |
|---|---|---|
| Disease population | People living with the condition, or developing it during a defined period | Eligibility for the proposed drug |
| Label-eligible population | People meeting the approved or assumed indication | Clinical suitability, access or willingness to start |
| Clinically suitable population | Eligible people considered appropriate for treatment after clinical assessment | Successful referral, funding or treatment initiation |
| Reachable population | Suitable patients who can realistically access treatment within the modeled period | Sales for an individual product |
| Product-treated population | Patients expected to receive the drug after adoption and competitive allocation | A full year of revenue from every patient |
Terms such as total addressable market, serviceable market and obtainable market can be useful shorthand. Define them anyway. A company presentation may describe a broad disease opportunity as its addressable market, while an investor uses the same phrase for reimbursed, treatment-ready patients. Neither number is comparable until its boundaries are clear. Keep the patient counts visible before attaching dollars. Otherwise, a large revenue figure can conceal an uncertain patient estimate, an ambitious price assumption, or both. For a pipeline drug, mark the indication as an assumption rather than an established regulatory outcome.
Choose prevalence, incidence or a patient-flow model
Prevalence measures people living with a condition at a defined time or over a defined period. Incidence measures new cases arising during a period. In cancer statistics, prevalence can include people diagnosed years earlier who remain alive, rather than only those currently requiring drug treatment; the National Cancer Institute’s definitions of cancer statistics distinguish these measures. A prevalence count therefore needs further interpretation before it becomes a treatment denominator. People with a history of cancer, patients with active metastatic disease and patients starting a new treatment line are not interchangeable groups.
Use a prevalence-based starting point when modeling a continuing pool of patients who could receive chronic treatment. Use incidence or treatment-entry flows when the opportunity depends on new diagnoses, relapses or progression into an eligible setting. Some forecasts need both: an existing pool at launch and new entrants afterward. Keep those cohorts separate. Adding annual incidence to a prevalence estimate that already covers the same period can count patients twice. For recurring treatment, track patients who remain on therapy across years; for a one-time intervention, model the untreated eligible pool and its depletion. The commercial pattern of gene therapies, for example, requires a different patient-flow structure from indefinite maintenance treatment.
Check whether the source reports patients, diagnoses or disease events. Cancer incidence counts can include multiple primary cancers in one person and generally exclude recurrences. Age-adjusted incidence rates also use a standard population structure, rather than the actual age mix of the population being sized. Those distinctions appear in the NCI methodology for cancer incidence rates. Prefer published case counts when appropriate. If counts must be calculated from rates, use matching population denominators and age-specific rates where available; do not simply multiply an age-adjusted rate by the national population and treat the result as an observed patient count.
Build an evidence base with matching denominators
Every material input needs a definition, geography, observation period and source date. Record those alongside the number, rather than leaving them in a folder of downloaded papers. A prevalence study may measure all disease stages, while a hospital dataset covers advanced cases referred to specialists. A claims analysis may identify diagnosed and insured patients, rather than everyone living with the disease. These datasets can answer different questions without contradicting one another. The practical approach in researching biotech through papers, registries and filings is to trace each estimate back to the population actually measured.
Use alternative datasets to test the model, not to assemble the largest possible opportunity. Compare a population-based estimate with observed diagnoses, specialist caseloads or treatment activity where suitable evidence is available. Investigate large gaps before averaging the numbers. They may reflect underdiagnosis, different disease definitions, referral patterns or overlapping records. For an investor presentation’s patient estimate, find the underlying study and check whether the company has added assumptions that the study did not establish. Where evidence remains weak, use an explicit range. A decimal place does not improve an uncertain denominator.
Translate the indication into an eligibility funnel
Separate the proposed label from trial enrollment criteria
For an approved drug, start with the indication and relevant patient-selection instructions in the prescribing information. For an investigational drug, build a base case around the population supported by the development program, then model broader uses separately. Approval remains uncertain. An initial indication could differ from the company’s long-term ambitions, so the drug approval process belongs in the assumptions behind the market model. Do not count every planned expansion as though it were already available at launch. Each expansion needs its own evidence, timing and patient definition.
Trial eligibility is informative, but it is not automatically the eventual commercial label. Studies may impose enrollment restrictions to manage safety, reduce variability or answer a focused research question. Treating every exclusion as a permanent commercial restriction can understate the opportunity. Ignoring those exclusions can overstate how broadly the evidence applies. Keep separate columns for trial criteria, formal label requirements and expected prescribing practice. The distinctions covered in clinical trial design and patient selection help explain why the enrolled population and the commercially treatable population may differ. Any adjustment between them needs a stated rationale.
Apply biological and clinical filters to the correct group
Kimmtrak illustrates how a disease headline can conceal several eligibility filters. Its FDA-approved indication covers HLA-A*02:01-positive adults with unresectable or metastatic uveal melanoma, with patient selection based on a positive genotyping test. These requirements appear in the FDA prescribing information for Kimmtrak. A market model cannot start with all melanoma patients and apply a single adoption percentage. It must identify the relevant melanoma subtype, disease setting, adult population and marker-positive group. Testing access then affects how many eligible patients can actually be identified and considered for treatment.
Use conditional percentages throughout this funnel. The marker-positive proportion among all diagnosed patients may differ from the proportion among patients with advanced disease or previous treatment exposure. Apply each percentage to the subgroup it actually describes. When two criteria overlap, estimate their intersection rather than assuming independence. The same discipline prevents repeated exclusions: a comorbidity might already be reflected in a clinical suitability estimate, so subtracting it again would shrink the market twice. Label each row with both its numerator and denominator. “Sixty percent eligible” is incomplete; “60% of marker-positive patients in the target stage meet the treatment-line requirement” can be audited.
A worked example: from 100,000 patients to 756 starts
Consider a hypothetical drug for a defined subgroup of a chronic disease in one country. The starting estimate is 100,000 people living with the condition on the launch date. The model narrows that population through diagnosis, indication requirements, clinical suitability and practical access. Its final row estimates starts on this particular drug during the first 12 months from that opening cohort. It is not an annual incident population, and it excludes patients who become newly eligible after launch. Every percentage below is illustrative and conditional on the preceding row. Real estimates would require evidence matching each subgroup.
| Step | Assumption applied | Patients remaining |
|---|---|---|
| People living with the disease at launch | Starting population | 100,000 |
| Diagnosed | 70% of the disease population | 70,000 |
| Within the intended disease stage | 40% of diagnosed patients | 28,000 |
| Required biomarker present | 25% of patients in that stage | 7,000 |
| Meets the treatment-line requirement | 60% of the preceding subgroup | 4,200 |
| Clinically suitable after assessment | 80% of diagnosed, label-eligible patients | 3,360 |
| Practical access within the first year | 75% of clinically suitable patients | 2,520 |
| Starts this drug during the first year | 30% of reachable patients | 756 |
Modeled first-year starts from the opening cohort = 100,000 × 70% × 40% × 25% × 60% × 80% × 75% × 30% = 756.
The model produces several useful numbers. There are 4,200 diagnosed patients meeting the assumed indication requirements, 3,360 judged clinically suitable and 2,520 reachable within the modeled year. The forecast of 756 starts sits below each of them. Its final 30% assumption combines treatment initiation on this product and competitive allocation over that period; it should not receive another unexplained adoption discount. A more detailed forecast could split that row into willingness to start treatment, prescriber adoption and product share. That is worthwhile only if the components can be estimated without counting the same barrier repeatedly.
This launch cohort also needs a time dimension. During the year, some untreated patients may progress beyond eligibility or otherwise leave the pool, while newly diagnosed patients may enter it. Either build those movements directly into the model or state that the first-year initiation assumption captures attrition from the opening cohort. Add newly eligible cohorts separately. Without that distinction, a spreadsheet can repeatedly apply the same start percentage to the original pool, even after many of those patients have already received treatment. The arithmetic will still work. The patient accounting will not.
Distinguish clinical eligibility from access to treatment
Leqembi provides another example of why a broad disease count is insufficient. Its prescribing information states that treatment should begin in patients with mild cognitive impairment or mild dementia stage Alzheimer’s disease and requires confirmation of amyloid beta pathology before initiation. It also calls for baseline and scheduled MRI monitoring. These requirements are set out in the FDA prescribing information for Leqembi. The market-sizing implication is a series of diagnostic and care requirements, rather than a direct conversion of all Alzheimer’s disease prevalence into potential starts. This is an investment-model observation, not a treatment recommendation.
Coverage adds another layer. The Medicare national coverage determination for antiamyloid monoclonal antibodies includes clinical eligibility requirements and coverage with evidence development; for drugs approved on direct measures of clinical benefit, study data may be collected through a registry. The CMS national coverage determination for Alzheimer’s antiamyloid therapies sets out those conditions. A model should therefore distinguish a patient who meets the drug’s clinical requirements from one whose care pathway also meets the applicable coverage requirements. Neither status alone establishes that treatment will begin during the forecast year.
Do not hide all access barriers inside “market penetration.” Separate coverage, referral, diagnostic completion and treatment capacity where the evidence permits. A patient awaiting testing represents a different forecasting issue from a patient whose insurer has declined funding. The former may enter treatment later; the latter may require a change in coverage or affordability. Keep the detailed economics in the pricing, reimbursement and market access model, but bring the resulting patient-access assumptions into the market forecast. State which barriers delay treatment and which remove patients from the forecast period altogether.
Model treatment capacity and the launch backlog
A clinically eligible population can exceed the number of patients a treatment system can process. Build a capacity cross-check from active centers, treatment slots, diagnostic throughput or another appropriate operating unit. Use the tightest relevant constraint, rather than assuming each capacity estimate represents an independent source of patients. If diagnosis can process 5,000 candidates but treatment centers can start only 1,000 patients, the larger figure does not determine starts. Product supply may impose a separate ceiling; the relationship between available doses and patient throughput belongs alongside biotech manufacturing capacity and production constraints. All capacity figures should match the same geography and period as the patient forecast.
The opening eligible pool is a backlog, not automatically a recurring annual opportunity. Model how quickly it can be assessed and treated, then track what remains. For a hypothetical one-time therapy with 12,000 untreated eligible patients and 1,500 starts per year, the backlog alone represents eight years of starts under a static calculation. That calculation assumes no new entrants, deaths, eligibility changes or competing treatments. It is a boundary check, not a forecast. For launch planning, build annual movements through referral, assessment and treatment, using the principles in drug launches and adoption to distinguish demand from delivery.
Allocate patients across products without double counting
A patient can be eligible for more than one treatment. That does not mean every manufacturer can claim the same patient as an exclusive commercial opportunity. Define whether the new drug replaces an existing therapy, is added to it or is used later. Product shares in a mutually exclusive treatment setting must reconcile to the same denominator. Where combinations or treatment sequences allow several products to earn revenue from one person, model the actual courses or periods of exposure. The patient overlap is then legitimate, but it must remain visible. A focused assessment of competing treatments and standards of care supplies the assumptions; the market model applies them.
Keep expansion opportunities separate too. Earlier treatment lines, pediatric use, another disease subtype and a new country each need an incremental population calculation. Before adding them, identify patients already counted elsewhere. Two indications may overlap, or the same person may move between treatment settings during the forecast. Maintain a patient-level concept even when the spreadsheet operates with cohorts. For a new indication, the useful question is how many additional treatment starts or paid treatment months it creates, not how large its headline disease population appears. This also prevents a portfolio forecast from becoming the sum of overlapping presentation slides.
Convert treatment activity into net revenue
Patient starts and revenue require different calculations. For ongoing treatment, forecast paid treatment months or doses, accounting for start dates, persistence, interruptions and discontinuation. For a fixed course, estimate courses delivered and net revenue per course. For a one-time intervention, estimate completed administrations and the relevant revenue assumptions. List price should not stand in for realized net revenue without an adjustment. Keep the price model separate from the patient model so that an apparent change in market size can be traced to patient volume, treatment exposure or price. That separation makes the forecast easier to update when commercial evidence arrives.
In the hypothetical funnel, 756 starts generating an average of nine paid treatment months at $8,000 net revenue per month would produce $54.432 million across those treatment months. That is cohort revenue, not necessarily revenue in the first calendar year. A patient starting on July 1 contributes only six months before year-end, assuming uninterrupted treatment through December. A monthly cohort schedule resolves this timing issue. If the nine-month assumption already reflects discontinuation and missed treatment, do not subtract those effects again. Likewise, patients continuing from earlier cohorts must be included once, rather than repeatedly added as new starts.
Cohort revenue = patient starts × average paid treatment months per start × net revenue per paid month.
Finally, distinguish product-market revenue from revenue attributable to the company being valued. A partner may hold commercial rights in some territories, or the developer may receive royalties rather than book product sales. The patient opportunity can remain unchanged while the company’s economics differ substantially. Apply territory rights and deal terms after constructing the relevant product-sales forecast, using the framework in valuing biotech licensing payments and royalties. Do not present a partner’s total product sales as the developer’s revenue, or treat the whole addressable market as a company-owned asset.
Test geography, assumptions and implied market share
Build countries or commercially similar regions separately before combining them. Differences in population age, diagnosis, treatment practice, launch timing and access can change the patient opportunity. A US eligibility fraction should not automatically be exported to another country, and a US net price should not price every patient globally. Use a consistent reporting currency and state the exchange-rate assumption. Where evidence is thin, a broad regional range is more defensible than a country-by-country table filled with unsupported precision. Separate an approved local opportunity from countries whose regulatory or reimbursement decisions remain pending.
Sensitivity testing should start with patient assumptions. In the worked example, changing diagnosis from 70% to 50%, with all other conditional assumptions held constant, reduces starts from 756 to 540. Raising diagnosis to 90% produces 972 starts. That is a one-variable test, not a prediction: in practice, patients found through broader diagnosis may have a different stage or biomarker mix. Build coherent downside and upside cases around such relationships. The methods used in testing valuation assumptions help identify whether uncertainty sits mainly in diagnosis, eligibility, access, adoption or treatment exposure.
Run the calculation backward as well. Divide a proposed sales figure by net revenue per treatment course or patient-year to obtain the treatment activity it implies. Compare that requirement with the eligible pool, realistic initiation rates and delivery capacity. A forecast can look modest against total disease prevalence while requiring most reachable patients to receive the product. Once the patient and revenue cases reconcile, they can feed into risk-adjusted pipeline valuation, where development probability, costs and timing are assessed separately. Market size is an input to that valuation, not its result.
Maintain a market model that evidence can change
Keep an assumption register rather than a single permanent market-size number. For each major input, record its definition, source, date, range and the next evidence that could change it. A label decision may alter the eligible population. Better diagnostic data may revise the diagnosed pool. Launch evidence may show that referral or treatment capacity is the binding constraint. Update the affected rows and preserve the earlier version so that changes in the investment case remain traceable. A smaller revised opportunity can be more useful than an unchanged large estimate whose original assumptions no longer fit the evidence.
- Can every percentage be matched to the population it describes?
- Are opening patients, new entrants and continuing treatment cohorts separated?
- Are eligibility, access and product adoption counted only once?
- Does the revenue forecast reconcile with paid treatment activity and commercial rights?
Use those checks whenever new evidence arrives. If a company raises its market estimate, identify the row that changed and the evidence supporting it. If projected sales rise without a change in patients, exposure, net price or commercial rights, the model is missing an explanation. Keep unsupported expansion opportunities outside the base case until their assumptions can be stated plainly. The working document should show what must happen for patients to reach treatment, how many can do so within the forecast period, and which unresolved assumption has the greatest effect on the investment case.