Change in stats when setting GroupOrder

[ManuelaS]

Hi,

I came across some unexpected behaviour when passing the "GroupOrder" flag. Specifically, the p-value differs depending on the order and in the last example below is a complex number. The GroupOrder flag worked as expected when testing it with the example datasets or other grouping variables of this dataset. I suspect the problem is related to convergence (the groups "appendix" and "others" have few patients with no events)

`
% Data from https://bitbucket.org/manuela_s/translating_network_biomarkers_into_the_clinic/src/master/data/clinical_data.csv
data = readtable('clinical_data.csv');
% Drop patients with missing censoring information
data.is_censored_overall_survival = categorical(data.is_censored_overall_survival);
data = data(~isundefined(data.is_censored_overall_survival), :);

% Simple call generates correct Kaplan-Meier plot and reports stats
[p, fh, stats] = MatSurv(...
data.overall_survival_months,...
data.is_censored_overall_survival=='no',...
data.tumour_site)

% Specifying "GroupsToUse" and including all the groups in the same order
% as default (NoOp) produces the expected Kaplan-Meier plot and stats
[p, fh, stats] = MatSurv(...
data.overall_survival_months,...
data.is_censored_overall_survival=='no',...
data.tumour_site,...
'GroupsToUse', {'appendix', 'distal', 'others', 'proximal', 'rectal'})

% Specifying "GroupsToUse" and including all the groups but in a different order
% than default produces the expected Kaplan-Meier plot, but reports
% different stats (and a complex number p-value in this case)
[p, fh, stats] = MatSurv(...
data.overall_survival_months,...
data.is_censored_overall_survival=='no',...
data.tumour_site,...
'GroupsToUse', {'appendix', 'proximal', 'distal', 'rectal', 'others'})
`

[aebergl]

Hello,
Thanks for taking time to find this bug! It turns out that the calculations of the log rank p-value do not work when there are two or more groups with no events. I have added a error checking for the groups that warns about this condition but still displays the KM-plot. I also by default remove groups with less than two samples. I have also added a so one can easily merge Groups with a multi-level cell structure as GroupsToUse input variable.
Your example would be fixed using:
[p, fh, stats] = MatSurv( data.overall_survival_months,data.is_censored_overall_survival=='no', data.tumour_site, 'GroupsToUse', {{'appendix+others','appendix','others'}, 'proximal', 'distal', 'rectal'});