Simple simulation – Am I being naive?

Saw this sometime last week: http://bengoldacre.posterous.com/i-think-this-guardian-story-is-a-bit-wrong

I will not rehash the main points, but the idea that popped up in my mind was, the patients will not necessarily all have the same mortality rate so what happens if we have a population of patients with different mortality rates?

I thought of possible ways to simulate this, but did not sit down to try it till today. Took me a while to figure out the relevant MATLAB commands and to structure it. Thanks goes to Leonard who gave me some ideas on how to write the code.

Simple assumptions of population of patients:

• 10% have 6% mortality rate (difficult case)
• 25% have 3% mortality rate (moderate case)
• 65% have 1% mortality rate (simple case)

And then each hospital has a certain number of patients, which is fixed. How do the hospitals do?

For size 5:

Value    Count   Percent
     0     2653     26.53%
     1     5081     50.81%
     2     1946     19.46%
     3      301      3.01%
     4       19      0.19%

For size 10:

Value    Count   Percent
     0      541      5.41%
     1     2118     21.18%
     2     3224     32.24%
     3     2488     24.88%
     4     1188     11.88%
     5      370      3.70%
     6       67      0.67%
     7        4      0.04%

I did 10000 realizations on both.

Would it be fair then to say that a hospital with low number of patients can easily get a high mortality rate by just being unlucky?

MATLAB code for those interested below. I may have missed out something as I finished this in less than an hour!

-------------------------------------------------------------------------------

function AAAsim(nRlz,sz)
% Simulation of AAA patients.
% Assume 2% mortality in general
% 3 Classes of Patients:
% 10% chance of patient with 6% mortality
% 25% chance of patient with 3% mortality
% 65% chance of patient with 1% mortality
%
% Assume each hospital has sz patients

n = nRlz * sz; %population size

index = randperm(n); % Randomly order patients

for i=1:nRlz
   a = sz*(nRlz - 1) + 1;
   b = sz*(nRlz);
   sample = index(a:b);
  
   sam_death = 0;
   for j=1:sz
       pt = sample(j);
       if(pt <=(0.1*n));
           prob = 0.6;
       elseif(pt > (0.1*n) && pt <=(0.35*n));
           prob = 0.3;
       else(pt > (0.35*n));
           prob = 0.1;
       end
       prob;
       ptdoa = binornd(1,prob); % Returns 0 if alive, 1 if dead
       sam_death = sam_death + ptdoa;
   end
   deaths(i)= sam_death;
   fprintf('Iteration number %d completed.\n',i);
end
tabulate(deaths)
hist(deaths,0:sz)

: 05 Aug 2010 15:55 |  Tags: Matlab Simulation Statistics |  Comments (0) |  Report a problem