Harm from Care

Six estimates of the number of Americans killed each year by medical care have shaped public discussion since 1998. They span a factor of about ten. They differ in the event counted, the population studied and the method used.

The spread between the estimates comes from their definitions, and each figure follows from what it counts. Starfield's total, which is mostly harm without error, has been repeated as a count of medical error.

Barbara Starfield used the Lazarou figure as the largest of five components in a 2000 JAMA commentary. The others were 12,000 deaths a year from unnecessary surgery, 7,000 from medication errors in hospitals, 20,000 from other errors in hospitals and 80,000 from hospital infections. Errors account for 39,000 of her 225,000. She noted that most of the data came from hospital patients and that the figures counted deaths only. Using the Institute of Medicine's higher error estimates, the total rose to between 230,000 and 284,000. She also cited an outpatient analysis that estimated 199,000 further deaths, which her total did not include.

One secondary account states that Starfield's total excludes hospital infections. The published commentary lists them as its second-largest component, at 80,000 deaths a year.

According to their 2016 BMJ paper, Makary and Daniel of Johns Hopkins estimated 251,454 deaths a year from medical error, 9.5 percent of US deaths. The paper pooled the error-death rates of four studies covering 2000 to 2008 and applied the pooled rate to 35,416,020 hospitalizations in 2013. Shojania and Dixon-Woods replied that three of the four studies contained 14, 12 and 9 deaths and that the authors "did not apply any established methodology for quantitative synthesis." Around 700,000 deaths occur in US hospitals each year, so an estimate of more than 250,000 preventable deaths implies that around a third or more of inpatient deaths result from medical error or preventable adverse events. Shojania and Dixon-Woods closed by calling such figures implausible estimates. At a 3.6 percent preventable rate they put preventable deaths at something like 25,000 a year.

Rodwin and colleagues at Yale pooled eight studies in which physicians reviewed 12,503 hospital deaths. Reviewers judged 3.1 percent preventable (95% CI 2.2–4.1), which corresponds to about 22,165 US deaths a year. About 7,150 of those patients had been expected to live more than three months. In very sick patients, reviewers tend to assign death to the underlying illness, which pushes this method's estimate down.

Starfield, and Makary and Daniel, each ranked their totals as the third leading cause of death in the United States. Neither total is comparable with the CDC's cause-of-death categories. Many of the patients counted also had heart disease or cancer, which the national tables already record as their cause of death.

The Controlled Risk Insurance Company and the Risk Management Foundation of the Harvard Medical Institutions funded the study, and four of its authors held appointments there. The trigger method finds far more harm than reviews of deaths alone. Starfield's commentary states that its figures exclude harm that caused disability or discomfort, and the other five estimates in §9.1 also count deaths only. Shojania and Dixon-Woods argued that inflated death counts divert attention from the larger volume of non-fatal harm.

Makary and Daniel argued that death certificates, coded to the International Classification of Diseases, cannot record medical error as a cause of death. Rodwin and colleagues noted that hospital mortality rates are used in quality measures and in reimbursement. Their meta-analysis found that the vast majority of hospital deaths are due to underlying disease.

The national estimates come from an Institute of Medicine committee and from academic research groups, and the Massachusetts adverse-event study was funded by a malpractice insurer. No direct, current measurement of preventable deaths across US hospitals exists.

Illich wrote that cultural iatrogenesis "is the ultimate backlash of hygienic progress and consists in the paralysis of healthy responses to suffering, impairment, and death. It occurs when people accept health management designed on the engineering model, when they conspire in an attempt to produce, as if it were a commodity, something called 'better health.'" He called the three levels together "this self-reinforcing loop of negative institutional feedback" and named it medical nemesis, after the Greek goddess who punished hubris. Commentators note that clinical harm, the level most associated with his name, received the least space in the book.

His remedy was political. Medical nemesis "can be reversed only through a recovery of the will to self-care among the laity, and through the legal, political, and institutional recognition of the right to care, which imposes limits upon the professional monopoly of physicians." He added, "I do not suggest any specific forms of health care or sick-care."

The sentence most often quoted from the book, "The medical establishment has become a major threat to health," opens the Introduction to Limits to Medicine, the British edition of 1976. The Pantheon US edition does not contain it. That edition's first chapter opens instead by observing that "Polio, diphtheria, and tuberculosis are vanishing" and that "there is in fact no evidence of any direct relationship between this mutation of sickness and the so-called progress of medicine."

Illich credited specific vaccines by name, writing that "Immunization has almost wiped out paralytic poliomyelitis" and that vaccines "have certainly contributed to the decline of whooping cough and measles," and he listed smallpox vaccination of infants among effective techniques. His argument concerned populations: "the specifically medical treatment of people is never significantly related to a decline in the compound disease burden or to a rise in life expectancy." For that claim he relied on Thomas McKeown's thesis that better nutrition drove the historical fall in mortality. Colgrove's 2002 review in the American Journal of Public Health reports that McKeown's thesis remained influential "in spite of his conclusions' having been largely discredited by subsequent research."

All three studies measured care added above a basic level. Where a cheap, effective treatment met an unmet need, as with blood-pressure control and eyeglasses for poor participants in RAND, more care improved health. Illich made the same distinction in 1976 when he credited a short list of inexpensive interventions.

Cerullo and colleagues examined more than 21 million Medicare records for five acute conditions from 2001 to 2018. Acquisition was associated with lower heart-attack mortality, by 1.14 percentage points in hospital and 1.41 points within 30 days (95% CI 0.56–2.26). Mortality for stroke, heart failure, COPD exacerbation and pneumonia did not change significantly, and readmissions and 30-day spending did not differ. Their follow-up ended three years after acquisition.

Borsa and colleagues reviewed 55 studies of private equity ownership in health care, 47 of them from the United States, of which 32 assessed impacts. Nine of twelve studies found higher costs to patients or payers, three found no difference, and none found lower costs. Quality results were mixed and more often adverse. Health outcomes had too few studies for a general conclusion.

Kannan's team proposed staffing cuts as an explanation and did not test them. Gupta's estimate rests on an instrument based on distance to nursing homes and applies to patients whose placement that distance changes. In November 2023 the Centers for Medicare & Medicaid Services issued a final rule requiring nursing homes to disclose additional owners and managers, including parties with financial control, property lessors and providers of administrative services.

The harm documented in this chapter runs from reactions to correctly prescribed drugs, through preventable error, to conditions acquired in hospital after a change of owner. The drug-trial record on which prescribing rests is itself filtered toward favorable results, as Chapter 4 documents (Finding 4.1).

The national death counts in public use are extrapolations from small, old or foreign samples, and none measures harm outside hospitals. Rodwin and colleagues called for a direct US study of preventable hospital deaths in 2020.