Product Field Work All builds
BuildLegible

What predicts a drug shortage?

The usual answer is “too few manufacturers.” Joining four FDA datasets shows that isn’t what the data says. Two other signals do, and in a one-year backtest they were visible well before the shortage notice.

The problem

When a drug goes short, a hospital pharmacy finds out from a wholesaler, then spends days finding substitutes, rewriting order sets and warning clinicians. The facts that might have warned them sit in separate FDA datasets that don’t reference each other: the shortage list, the drug directory and the recall reports.

–shortage notices listed as current on FDA’s list today, including discontinuations
8.6Mextra hours U.S. hospitals spend managing shortages each year, per a 2019 Vizient survey
$359Ma year in labor alone, from the same survey; it excludes costlier substitutes and delayed care

The cost isn’t only hours. In the same survey, 38% of hospitals reported at least one medication error tied to a shortage in six months.

How I measure the problem
  1. How much more often medicines with a given trait are short, tested for every popular theory
  2. Whether a signal known a year ago predicted the next year’s shortages, a backtest with nothing from the future
  3. Hours a hospital could save by acting before the notice, with assumptions you can change
Why I chose it

It’s a question that needs data joined, not just read, and I know how drugs are made and regulated well enough to spot a join that looks right and isn’t. It also let me test a belief almost everyone repeats.

The ideaJoin the shortage list, the drug directory and recall reports on exact FDA identifiers. Test each popular theory against the result, keep only what holds, and backtest it.

Sources: openFDA drug shortages, NDC directory, Drugs@FDA and enforcement reports, downloaded September 26, 2026; Vizient, “Drug shortages and labor costs,” June 2019.

Test a theory

What we’d have known a year ago

Signals as of September 26, 2025, for medicines that had never been on the shortage list. The outcome is whether any shortage or discontinuation notice was posted in the following year.

The watch list

Injectable medicines not on the shortage list today whose own record, or a maker’s record, shows a manufacturing-quality recall in the past three years. A signal, not a prediction.

What the warning is worth

For one hospital pharmacy that gets an early warning on the shortages this signal would have caught. The catch rate is measured in the backtest above.

–of pharmacist time saved a year, per hospital
–hours a year back, net of reviewing the watch list
–shortages a year seen coming, of the ones this hospital manages

    From public data

    Median pay for pharmacists, BLS, May 2024.

    Private industry, BLS ECEC, June 2026.

    Two-thirds of Vizient respondents managed at least 20 in six months, so 40 a year is a floor.

    8.6 million hours across roughly 6,000 hospitals is about 1,430 hours each, or about 36 per shortage at 40 a year.

    Assumptions you can change

    No public benchmark exists for these. They are starting points, not findings.

    What it means for the people

    Today, pharmacy buyers and shortage coordinators react: a notice arrives, and the scramble starts. Here is how a year of shortage work shifts.

    Today
    Reacting to every shortage after the notice
    With the signal
    Watch list
    Still reacting
    Freed by planning ahead

    Buyers plan instead of scrambleA flagged injectable is a reason to confirm a second source or add a little safety stock, weeks before anyone else is looking.
    Pharmacists prepare substitutions calmlyOrder sets and dosing guidance for the likely alternative can be ready before the first backorder, which is where errors come from.
    Every flag has a receiptEach medicine on the list shows the recall behind it, with date, firm and reason, so a buyer can judge it in a minute.
    What disappears, and what doesn’t

    Some of the scramble goes away for the shortages seen coming. Two-thirds still arrive without warning, and deciding what to stock, what to substitute and how to tell clinicians stays with the pharmacy.

    How it works

    Four FDA datasetsShortages, the NDC directory, Drugs@FDA applications, and recall enforcement reports.
    Exact joinsA medicine is its ingredient codes plus dosage form and route. Recalls link by application number, or by maker name after normalization.
    Test, then backtestEach theory is checked against today’s list, then against what happened after a fixed date.
    A watch list with receiptsOnly the signals that held, each tied to the recall that triggered it.

    Evidence

    –of FDA’s shortage notices joined to an exact medicine, with no fuzzy matching
    –the rate at which flagged injectables went short in the backtest, against the base rate
    –recall reports tied to a maker, up from with an application number alone
    What I haven’t proven yet

    The backtest is one year and the counts are small, so treat the lift as a direction, not a precise number. The rule catches about a third of new notices and most flags never become shortages. Plant-level FDA inspection results would sharpen it, but they sit behind a registered API I haven’t connected. The outcome also counts discontinuation notices, which aren’t the same as a shortage. And compounders and repackagers, most of the unmatched recalls, aren’t manufacturers of record, so they’re deliberately left out.

    Choices

    Test the popular theory before building on it“Too few makers” is the obvious feature. The data says it points the wrong way at this level, so it isn’t in the rule.
    FDA identifiers over fuzzy drug namesIngredient codes, application numbers and dosage forms join exactly. Name matching is used only for makers, and only after strict normalization.
    A signal with receipts, not a risk scoreA buyer can weigh “sterility recall at this maker, March 2026.” A score of 0.73 gives them nothing to check.