Almost none of the change in US overdose deaths tracks the drop in opioid prescribing

Summary

Almost none of it. Opioid prescribing has fallen essentially every year since 2012 — from a peak of 81.3 prescriptions per 100 people to about 35 in 2024, a 56% drop. Over that same span overdose deaths did not fall in step; they rose 92% net, because people who had become dependent on prescription opioids increasingly shifted to heroin and then illicit fentanyl, whose supply and potency — not the prescribing rate — set the death toll. Correlating the two series' year-to-year changes (not just their levels, which are both trending and would show a spurious link) gives r = -0.17, R² ≈ 0.03: about 3% of the annual variation in overdose deaths lines up with the annual change in prescribing, and even that sliver runs the wrong way — years with the steepest prescribing declines were mostly years deaths rose fastest. The one exception is 2023-2024, when both prescribing and deaths fell together — but the leading explanation for that recent drop, per CDC and independent researchers, is a disruption in the illicit fentanyl supply, not a change in the pace of prescribing (which was decelerating, not accelerating, relative to the 2012-2019 period).

Step 1 — The two series, side by side

National opioid prescribing rate (CDC/IQVIA Xponent, retail pharmacy dispensing per 100 persons):

National overdose deaths, all intents (CDC WONDER multiple-cause-of-death, queried directly via AskAmerica's health.cdc_wonder_multiple_cause_of_death_1999_2024):

Prescribing peaked in 2012 and has declined every measured year since. Deaths kept climbing for another decade, peaking in 2021-2022, then fell in 2023-2024.

Step 2 — Why levels correlate but changes don't (and why the change correlation is the right test)

If you correlate the raw levels of the two series 2006-2024, you get r ≈ -0.94 — deaths and the prescribing rate look tightly (inversely) related. That is a classic two-trending-series artifact: both series have a strong, mostly monotonic trend over this span (prescribing down, deaths up), so almost any two long-trending series will show a strong correlation in levels regardless of whether they are causally connected. The question — how much of the change in overdose deaths tracks the drop in prescribing — is better answered by correlating year-over-year changes in each series, which strips out the shared trend and asks whether the two move together on a year-by-year basis. That correlation is r = -0.17 (R² ≈ 0.03), computed in Python from the two series above. Only about 3% of the year-to-year swings in overdose deaths coincide with the swings in the prescribing rate, and the sign is backwards from what a causal 'less prescribing → fewer deaths' story would predict.

Step 3 — What the prescribing decline did and didn't do to the death toll

• 2012 → 2019: prescribing fell 42% (81.3 → 46.8 per 100); overdose deaths rose 69% (45,135 → 76,087).
• 2019 → 2022 (peak): prescribing fell another 16% (46.8 → 39.5); deaths rose another 51% (76,087 → 114,658).
• 2012 → 2024, full span: prescribing down 56.5%; overdose deaths still up 91.7% net, despite the recent decline.
• 2022 → 2024, the recent decline: prescribing down another 10.4% (39.5 → 35.4) — a smaller, decelerating drop compared to 2012-2019 — while overdose deaths fell 24.5% (114,658 → 86,546), the sharpest two-year drop in the whole series.

Deaths specifically involving prescription (natural/semisynthetic) opioids tell the same story at finer grain: they rose from 3,442 in 1999 to a peak of 17,029 in 2017, then declined to about 13,026 in 2023 — a drop of roughly 4,000 deaths. That entire decline is smaller than one-tenth of the ~41,000-death net increase in total overdose deaths between 2012 and 2024, so even the piece of the mortality trend that prescription opioids can plausibly explain does not explain the aggregate rise.

Step 4 — What does track the change, per the research literature

The research consensus, corroborated across CDC surveillance reports and independent studies, attributes the two big features of the trend to factors other than prescribing policy:

• The 2013-2022 rise is attributed to the heroin/illicit-fentanyl transition: as prescription-opioid supply tightened, people already dependent shifted to heroin (the mid-2010s 'second wave'), then to much more lethal illicitly manufactured fentanyl (the 'third wave' from about 2013 onward), which by 2021 was involved in roughly 70,600 overdose deaths versus about 13,600 for prescription opioids alone.
• The 2023-2024 decline is attributed by CDC and by a 2026 University of Maryland/international research team to a disruption in the global illicit fentanyl supply chain (precursor-chemical restrictions reducing fentanyl potency/availability), not to any acceleration in prescribing reductions — the prescribing decline was in fact decelerating over this same window. One CDC finding that further undercuts a prescribing-driven story: the 2023-2024 decline was broad-based across cocaine, methamphetamine, heroin, methadone AND prescription-opioid-involved deaths together, which argues against a single drug-class-specific driver like the fentanyl-potency hypothesis alone, but still points away from prescribing as the mechanism, since prescribing itself did not change abruptly in 2023-2024.
• Naloxone distribution expansion and broader access to medications for opioid use disorder are also cited by CDC and outside researchers as contributors to the recent decline, entirely independent of the prescribing rate.

Caveats

• The two series come from different sources and slightly different definitions: overdose death counts here are CDC WONDER's multiple-cause-of-death series (a cause counts if listed anywhere among up to 20 causes, all intents included), which runs a bit higher than the provisional NCHS press-release total (e.g., 112,108 for 2023 here vs. 105,007 in NCHS's provisional data brief, and 86,546 for 2024 here vs. roughly 79,384 in the NCHS 2024 estimate) — direction and magnitude of the trend agree across both, but the exact death counts should not be treated as interchangeable with the NCHS headline figures.
• The opioid prescribing rate is CDC's published IQVIA Xponent-based dispensing rate; it is not carried in the AskAmerica connector's tables (confirmed via search_catalog — no opioid prescribing/dispensing-rate table exists in this warehouse), so those figures were read directly from CDC's own published pages (Opioid Dispensing Rate Maps, MMWR Vital Signs) rather than computed from a government dataset in this session. Years between CDC's published milestone points (2006, 2010, 2012, 2015, 2016, 2019, 2020, 2022, 2023, 2024) were linearly interpolated to build a full annual series for the correlation calculation — this could understate or overstate a mid-milestone inflection, but no published year in that window contradicts the milestone-to-milestone direction.
• Correlation, whether on levels or on changes, does not establish causation in either direction; the change-correlation calculation here is offered as a quantification of co-movement, with the causal account (fentanyl supply/heroin substitution, not prescribing) drawn from the cited CDC and academic literature, not derived from the correlation itself.