Tonsillectomy: bleeding and prediction

Tonsillectomy is a common operation, and in by far the majority of cases is carried out without any complications. One of concern is postoperative bleeding. In the first 24 hours it is an immediate consequence of operation, but it can occur after the immediate postoperative period and up to 10 days after operation. Tonsillectomy in countries like the UK is now complicated by the need for disposable equipment because of fear of spreading prion diseases.

Could postoperative bleeding be predicted by preoperative tests of abnormal coagulation? A meta-analysis tells us that this is unlikely [1], and has the additional benefit of giving good information on postoperative bleeding rates.


The review used wide searching strategies including the Cochrane Library to find studies examining tonsillectomy and bleeding, with an additional criterion of coagulation tests in people who bled postoperatively. It is not clear that the studies had to conduct preoperative coagulation tests in all patients.

Included studies were those reporting on people undergoing tonsillectomy and/or adenoidectomy, that were prospective, and had groups without concomitant illness. Information obtained was on the end point of bleeding with normal or abnormal coagulation tests.


There were no randomised trials, but four prospective studies with 3384 patients fulfilled the criteria, and six retrospective studies with 8988 patients were included for sensitivity analysis.

The bleeding rate was 3.4% (116/3384; 95% CI 2.8% to 4.0%) in prospective studies and 2.3% (207/8998; 95% CI 2.0% to 2.6%) in retrospective studies. Overall it was 2.6% (323/12.372; 95% CI 2.3% to 2.9%). Smaller studies were more variable (Figure 1). Bleeding was usually delayed, but this was not always specified.

Figure 1: Bleeding rate in tonsillectomy. Dark symbols prospective studies, light retrospective

In 323 patients who bled, abnormal coagulation tests were found in 24 (7.4%, 95% CI 4.6% to 10.3%), and identical rates were found in prospective and retrospective studies.


Readers will need to take this paper slowly. As best Bandolier can judge, of the 12,372 patients 323 bled, and coagulation tests are given on these only. The papers did not use universal preoperative screening. This is important because the paper reports sensitivity, specificity and positive and negative predictive values. For analysis after the event, these are unhelpful. We do not know how many people had normal or abnormal coagulation tests and did not bleed. It is a good example for critical appraisal of diagnostic tests, though.

But in reality the low rate of abnormal tests among the very few patients who did bleed absolves us from the need to think about this further. The results cannot get better with universal preoperative screening, and that is helpful in that it makes the result a pretty positive negative.

The most useful aspects of the study may lie in the light it sheds on variability of bleeding rates. Case mix and surgical competence will be important determinants, but one cannot get over the random play of chance. Two prospective and one retrospective study had bleeding rates of over 6%. The two prospective studies had about 100 and 200 patients respectively.

Suppose this was an audit? Would the immediate reaction be to:

  1. Blame the surgeon(s)?
  2. Blame the hospital?
  3. Blame the government?
  4. Blame the small sample size?

We need to have a much better handle on the last before we start thinking about one of the first three.


  1. P Krishna, D Lee. Post-tonsillectomy bleeding: a meta-analysis. Laryngoscope 2001 111: 1358-1361.