Atul Gawande's Complications: A Surgeon's Notes on an Imperfect Science is a mixed bag for rationalists. Written as a series of essays organized into three sections entitled Fallibility, Mystery, and Uncertainty, the book as a whole is of questionable value, but the sections need to be considered individually for their worth to be accurately recognized.
Fallibility examines a number of issues involving error and how it can be avoided in medicine, including but not limited to: how computers are better than humans at making diagnoses, the factors that are known to negatively influence human reasoning when applied to diagnosis and treatment, how and why doctors make errors, what happens when doctors "go bad", and ways that the system fails and succeeds at curbing irresponsible physicians. The section dealing with the field of anesthesiology is especially interesting, as Gawande recounts the changes that made it possible for the death rate due to general anesthesia to be reduced to a twentieth of its previous value midway through this century.
What is distressing is that Mystery and Uncertainty do not represent good examples of the principles that Gawande demonstrates he understands in Fallibility, supporting his assertions with vivid anecdotes instead of reasoning from principles. One story is especially obnoxious, as Gawande recounts how a "gut feeling" about a case of infection causes him to persuade a patient to undergo an extensive biopsy which reveals she has life-threatening necrotizing fasciitis, caused by the popularly-known "flesh-eating bacteria". He explicitly acknowledges that he had no logical reason for regarding the case as unusual, that all appearances indicated a routine cellulitis which was approximately three thousand times more likely than necrotizing fasciitis was - and that he had recently taken care of a patient who had died agonizingly from a massive infection of the flesh-eating bacteria. And yet the story is presented as an example of how to deal with uncertainty.
If a normal infection had been present, the incident would likely not have been recounted and possibly not even remembered with clarity by Gawande. We don't know - and likely Gawande doesn't know either - whether his hunches are more likely than not to be accurate. As he discusses himself earlier in the book, such hunches are known to be usually wrong, but they make vivid anecdotes - which renders them potent and dangerous lures for erroneous thinking.
On other problematic topic, he decries the relative rarity of autopsies and notes that they demonstrate that doctors are wrong about the cause of death and diagnosis of roughly one-third of patients who die while in care. Yet despite also noting that this fraction has not changed since the 1930s, he implies that the reduction is due to the unwillingness of doctors to trouble grieving family members rather than drawing the obvious conclusion that doctors don't utilize or care about autopsy results.
The book is worth reading, both to gain knowledge about the troubling subject of medical error and how it is and isn't reduced, and as an object lesson in how difficult it is for even intelligent and informed people to actually apply principles of proper reasoning.