Depends what your measure of sensitivity is. One possibility is to do a Monte-Carlo simulation and use the standard error on the mean as the measure. For example:
m = 1000;
Yfn = @(A,B,C) (-1.8*A + 0.025*B + 0.33*C + 0.0025*m)./(1 - 1.13*A + 0.011*B);
% Constraints - assuming the constants A, B and C are uniformly
% distributed between lower and upper limits.
N = 10^6; % Number of trials
A = pi/36 + rand(N,1)*(pi/6 - pi/36);
B = 20 + rand(N,1)*40;
C = 1.6 + rand(N,1)*3.4;
Y = Yfn(A,B,C); % N values of Y
Yav = mean(Y); % average value of Y
sigma = std(Y); % standard deviation
se = sigma/sqrt(N); % standard error
disp(' Yav se')
disp([Yav se])
Or use the standard deviation as the measure.
Or, if you want the sensitivity to each parameter separately, then do the same as the above for each separately keeping the other two fixed at their nominal values.
