The article is from The Journal of the American Medical Association: Is US
Health Really the Best in the World?
Barbara Starfield, MD, MPH, JAMA. 2000;284:483-485.

Dr. Starfield estimated that in the US, there are 12,000 deaths/year from
unnecessary surgeries, 7000 deaths/year from medication errors in
hospitals, 20,000 deaths/year from other errors in hospitals, 80,000
deaths/year from nosocomial infections in hospitals, and 106,000
deaths/year from nonerror, adverse effects of medications.

The estimated deaths are for the most part not due to "medical blunders."
The total for the largest group, nonerror, adverse medication effects, is
an extrapolation based on limited reviews of adverse drug reaction forms
(the forms healthcare workers fill-out when a drug reaction or adverse
effect is suspected).

These statistics are important to consider, but they must be considered in
context. There's a very high correlation between people who are already
sick and people who take medication, so most of those who die of
iatrogenic causes are, by definition, already sick. 'Stong' (effective)
medicines often have strong adverse effects. Medications treat sickness,
and sickness kills people. Even though many people die as a result of
medications, many more of those sick people would die without those
medicines than die from their adverse effects.

Many people would die without medical intervention, and some do die
anyways inspite of or because of heroic and desperate attempts to save
them. It's tragic when anyone dies from a medical error, but it's also
tragic if anyone dies from an illness that could be treated. Many of those
treatments, however, entail risks. We could easily decrease the numbers of
deaths from medical errors by simply not treating the sickest patients,
but then those people would be much more likely to die from their
illnesses.

The second largest group, the nosocomial (hospital-acquired) infection
deaths, does contain many potentially preventable deaths, but here again,
many of those infected are already very ill and often immunocompromised.
For example, many cancer therapies weaken the immune system and leave
patients vulnerable to infection. It would be easy to completely eliminate
such deaths by simply not treating the cancers, but then those cancers
would almost certainly be fatal. That would improve "medical error"
statistics, as there would be fewer medication related deaths, but more
people would die in the end then are now saved.