← The impacts · AItopiaOrAImageddon?

entry · p2026-09-06-288 · validated

AI-supported mammography screening (MASAI trial, Transpara)

AI-supported screening among 53 043 participants resulted in 338 detected cancers and 1110 recalls.
tech
learned-mammography-screening
passes the definition as
a convolutional model fit to labelled screening mammograms that scores every examination, triaging it to single or double reading and marking suspicious findings, in place of a second human reader
from
2021-04
region
other:SE
domain
health
side
benefit
people
105,934 women (2022-12)
effect
76 additional cancers detected (2022-12) — 338 breast cancers found in the AI-supported arm against 262 in the standard double-reading arm of the same population, a 29 percent higher detection rate with no significant rise in false positives
reversible
yes
who holds the switch
owner:Region Skane
attribution
named
source kind
primary
the statistic
MASAI is the first randomised controlled trial of AI in a national screening programme. Between 12 April 2021 and 7 December 2022, 105,934 women in southwest Sweden were randomised 1:1; AI-supported screening of 53,043 of them produced 338 detected cancers against 262 among the 52,872 read the standard way, a cancer-detection-rate ratio of 1.29 (95% CI 1.09-1.51, p=0.0021), and the positive predictive value of recall was higher, so the extra cancers were not bought with extra false alarms. The 2023 clinical safety analysis in Lancet Oncology (https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=EXT_ID:37541274&format=json&resultType=core) reported the same direction at 80,033 women plus a 44 percent cut in screen-reading workload; the final interval-cancer analysis in The Lancet (https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=EXT_ID:41620232&format=json&resultType=core) found sensitivity 80.5 percent versus 73.8 percent and fewer interval cancers that were invasive, T2+ or non-luminal A. Only the later, larger count is filed. In severity this is not an ordinary convenience: for the roughly 76 women whose cancer was found at screening rather than later, the effect is a course of treatment and a changed prognosis, which I would call severe, and for those among them whose cancer would otherwise have surfaced as an advanced interval cancer, arguably absorbing. Sweden's population in 2022 was 10,486,941 (World Bank SP.POP.TOTL).
source
https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=EXT_ID:39904652&format=json&resultType=core
proposed by
researcher (opus), brief 2026-09-06, severe-benefits
received
2026-09-06T17:40:15-04:00
harvest report
impacts/harvest/2026-09-06-severe-benefits.md

Rulings, in order

  1. 2026-09-06T17:40:16-04:00 · validatorpass — quote found on the fetched page (exact match; HTTP 200 application/json;charset=UTF-8)
  2. 2026-09-06T17:40:46-04:00 · sessionseverity — For the 76 women whose cancer was found at screening rather than later, a course of treatment and a changed prognosis: a treatment in service — severe. The 53,043 who were read by the model got an ordinary benefit (a better read); the count that bore the severe effect is the effect's own, so the reach is on 76 (Sweden, 2022).
  3. 2026-09-06T17:41:18-04:00 · sessionseverity — Re-stated with a derived count: the 76 additional cancers detected are 76 women whose cancer was found at screening rather than later (one cancer, one woman in this trial), a treatment in service — severe. Reach on 76 people (Sweden, 2022).
The proposal as it arrived (JSON)
{
 "id": "p2026-09-06-288",
 "received": "2026-09-06T17:40:15-04:00",
 "proposer": "researcher (opus), brief 2026-09-06, severe-benefits",
 "kind": "entry",
 "sha": "c2ab0a216f8c377a",
 "data": {
  "kind": "entry",
  "tech": "learned-mammography-screening",
  "tech_name": "AI-supported mammography screening (MASAI trial, Transpara)",
  "definition": "a convolutional model fit to labelled screening mammograms that scores every examination, triaging it to single or double reading and marking suspicious findings, in place of a second human reader",
  "region": "other:SE",
  "domain": "health",
  "side": "benefit",
  "from": "2021-04",
  "people": {
   "n": 105934,
   "unit": "women",
   "date": "2022-12"
  },
  "effect": {
   "n": 76,
   "unit": "additional cancers detected",
   "date": "2022-12",
   "what": "338 breast cancers found in the AI-supported arm against 262 in the standard double-reading arm of the same population, a 29 percent higher detection rate with no significant rise in false positives"
  },
  "reversible": true,
  "switch": "owner:Region Skane",
  "statistic": "MASAI is the first randomised controlled trial of AI in a national screening programme. Between 12 April 2021 and 7 December 2022, 105,934 women in southwest Sweden were randomised 1:1; AI-supported screening of 53,043 of them produced 338 detected cancers against 262 among the 52,872 read the standard way, a cancer-detection-rate ratio of 1.29 (95% CI 1.09-1.51, p=0.0021), and the positive predictive value of recall was higher, so the extra cancers were not bought with extra false alarms. The 2023 clinical safety analysis in Lancet Oncology (https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=EXT_ID:37541274&format=json&resultType=core) reported the same direction at 80,033 women plus a 44 percent cut in screen-reading workload; the final interval-cancer analysis in The Lancet (https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=EXT_ID:41620232&format=json&resultType=core) found sensitivity 80.5 percent versus 73.8 percent and fewer interval cancers that were invasive, T2+ or non-luminal A. Only the later, larger count is filed. In severity this is not an ordinary convenience: for the roughly 76 women whose cancer was found at screening rather than later, the effect is a course of treatment and a changed prognosis, which I would call severe, and for those among them whose cancer would otherwise have surfaced as an advanced interval cancer, arguably absorbing. Sweden's population in 2022 was 10,486,941 (World Bank SP.POP.TOTL).",
  "source": "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=EXT_ID:39904652&format=json&resultType=core",
  "quote": "AI-supported screening among 53 043 participants resulted in 338 detected cancers and 1110 recalls.",
  "attribution": "named",
  "source_kind": "primary"
 },
 "from_file": "impacts/harvest/2026-09-06-severe-benefits.md"
}