In the Harvard Medical Practice Study, a sample of 31,429 medical records of hos
ID: 3049687 • Letter: I
Question
In the Harvard Medical Practice Study, a sample of 31,429 medical records of hospital patients were reviewed to assess the frequency of medical malpractice. Two types of malpractice were identified:
1) An adverse event was defined as an injury caused by medical management (rather than by the underlying disease).
2) Negligence was defined as care that fell below the standard expected of physicians in the community.
An approximate 1% sample of records was reviewed on two different occasions by different review teams. The data were obtained in the table below.
Data
Adverse Events
Review Process B
Review process A 35 13
21 249
Negligence
Review Process B
Review Process A 4 9
12 293
1) Are frequencies of reporting of adverse events or negligence comparable in review process A (the original review) and review process B (the re-review)?
2) Can you assess the reproductibility of adverse events and negligence designations? Which seems to be more reproducible?
Explanation / Answer
now observed accuracy = (35+249)/(35+249+21+13) =0.89
expected accuracy =( ((35+21)*(35+13))/(35+249+21+13) + 249+13)*(249+21)/(35+249+21+13))/ (35+249+21+13) = 0.72
putting the values in the formula
we get kappa as (0.89-0.72)/(1-0.72) = 0.60
now for the other one , negligence using the same formula
now observed accuracy = (4+293)/(4+293+9+12) =0.933
expected accuracy = (16*13/318 + 302*305/318)/318 = 0.912
so kappa = (0.933-0.912)/(1-0.912) = 0.238
hence the kappa of adverse events is 0.6 as compared to 0.238 of negligecen events , hence adverse events has more reproducibility