| 1 |
M S S S t e s t |
| 2 |
=============== |
| 3 |
|
| 4 |
|
| 5 |
MSSStest is a program for implementing the method described in the |
| 6 |
article The Multiple Sclerosis Severity Score. R. Roxburgh S. Seaman |
| 7 |
et al. (2004), accepted for publication by Neurology. It calculates |
| 8 |
MSSS scores and uses these scores to test for differences between |
| 9 |
disease progression rates in different groups defined by genotype at |
| 10 |
some locus. A brief description of the method is given below. |
| 11 |
|
| 12 |
Introduction |
| 13 |
============ |
| 14 |
|
| 15 |
Suppose one is interested in determining whether the genotype at some |
| 16 |
locus of interest affects the rate of progression of disease of people |
| 17 |
with Multiple Sclerosis (MS). One method would be to calculate the |
| 18 |
mean EDSS score in patients with each of the genotypes and test whether |
| 19 |
they are significantly different. However, this approach is |
| 20 |
inefficient if patients have their EDSS assessed at different durations |
| 21 |
of disease, as patients who have had MS longer typically have higher |
| 22 |
EDSS scores regardless of their genotype. By adjusting for duration, a |
| 23 |
more powerful test can be developed. |
| 24 |
|
| 25 |
MSSS Method |
| 26 |
=========== |
| 27 |
|
| 28 |
Given a dataset of EDSS scores measured on patients at different |
| 29 |
durations of disease, a table is generated to convert ESSS scores to |
| 30 |
MSSS scores. This is done as follows. For each i (i=0, ..., 30), all |
| 31 |
patients having durations between i-2 years and i+2 years are ranked |
| 32 |
according to their EDSS scores. Suppose there are Ni such patients. |
| 33 |
Then, for each possible EDSS value (0, 1, 1.5, ..., 9.5), the average |
| 34 |
rank for that value is calculated. Suppose that for year i and an EDSS |
| 35 |
of j, this average rank is Rij. Rij is divided by Ni + 1 and then |
| 36 |
multiplied by 10, to yield a number between 0 and 10. This is the MSSS |
| 37 |
score for a person whose EDSS is measured as j when their duration is i |
| 38 |
years. This procedure is repeated for each value of i in turn. |
| 39 |
|
| 40 |
A patient having an MSSS of x progresses faster than 10x% of MS |
| 41 |
patients in the population and slower than (100-10x)%. So, for |
| 42 |
example, an MSSS of 5.0 means progressing at the median rate. A |
| 43 |
patient whose MSSS is 9.0 is a fast progressor, progressing faster than |
| 44 |
90% of patients. A patient whose MSSS is 1.0 is a slow progressor, |
| 45 |
progressing faster than just 10% of patients. |
| 46 |
|
| 47 |
Using data on 9892 patients from 11 (mainly European) countries, we |
| 48 |
derived an MSSS table. This is the `Global MSSS' table, which can be |
| 49 |
used to look up, for each patient in a study, the MSSS score that |
| 50 |
corresponds to their EDSS and duration. However, an alternative is to |
| 51 |
generate an MSSS table from any particular set of data being analysed. |
| 52 |
This is a `Local MSSS' table. |
| 53 |
|
| 54 |
Once MSSS scores have been assigned to patients (whether using the |
| 55 |
Global or a Local table), the Kruskal-Wallis test can be used to |
| 56 |
compare median MSSS in the different genotype groups. The |
| 57 |
Kruskal-Wallis test is similar to the ANOVA test, but is a |
| 58 |
non-parametric test, i.e. it does not assume normally distributed |
| 59 |
data. If there are only two genotype groups, the Kruskal-Wallis test |
| 60 |
is identical to the Wilcoxon test (also known as the Mann Whitney U |
| 61 |
test). Patients with duration 0 years are excluded from the test |
| 62 |
because EDSS assessments in the first year are not adequately |
| 63 |
predictive of later disease progression. |
| 64 |
|
| 65 |
The MSSStest program performs this Kruskal-Wallis test, having first |
| 66 |
assigned MSSS scores using either the Global MSSS table or a Local MSSS |
| 67 |
table generated from the data provided. We recommend that in nearly |
| 68 |
all cases, the Global table should be used. A Local table should be |
| 69 |
used only if the sample is large (>1000 patients) and the method of |
| 70 |
recruitment of patients is such that the distribution of EDSS |
| 71 |
conditional on duration is very different from that of the combined |
| 72 |
cohort of 9892 patients used to calculate the Global table. This could |
| 73 |
be, for example, if a study recruited only fast and/or slow |
| 74 |
progressors. |
| 75 |
|
| 76 |
P-values for the Kruskal-Wallis test can be obtained either by using an |
| 77 |
asymptotic (large sample) approximation or by an exact, permutation |
| 78 |
method. The asymptotic method should be fine for large samples. |
| 79 |
However, if the sample is small, ask MSSStest to also calculate a |
| 80 |
permutation p-value. In most cases, this should be very similar to |
| 81 |
that calculated by the asymptotic method. If they are very different, |
| 82 |
the sample is too small for the asymptotic approximation to be reliable |
| 83 |
and the permutation p-value should be used instead. |
| 84 |
|
| 85 |
Another application of the MSSS is to describe disease severity in a |
| 86 |
single dataset. If all patients entered into MSSStest are in a single |
| 87 |
dataset it will calculate the mean Global MSSS for this dataset but |
| 88 |
will not perform a Kruskal-Wallis test. The mean is calculated |
| 89 |
excluding patients with duration 0 years because EDSS assessments in |
| 90 |
the first year are not adequately predictive of later disease |
| 91 |
progression. |
| 92 |
|
| 93 |
Input files |
| 94 |
=========== |
| 95 |
|
| 96 |
MSSStest requires as input files: |
| 97 |
|
| 98 |
edss.txt |
| 99 |
This should be a tab or space delimited text file (ASCII) file. It |
| 100 |
should contain one row for each individual in the data set. Five |
| 101 |
columns are required. The first two columns are simply labels for the |
| 102 |
individuals in the data set. They are not used to perform the test. |
| 103 |
The first column is the family number and the second column is the |
| 104 |
number of the individual in the family. The third column is for the |
| 105 |
EDSS score. The fourth is for the disease duration at the time the |
| 106 |
EDSS was measured, rounded down to the next lowest whole year. The |
| 107 |
fifth is for the group to which the individual belongs. Groups are |
| 108 |
coded numerically and should be between 0 and 99. An example edss.txt |
| 109 |
file is provided. Note that no header row is allowed. |
| 110 |
|
| 111 |
global.dat |
| 112 |
This is the Global MSSS table: a matrix for converting EDSS scores into |
| 113 |
MSSS scores. It is not necessary to understand it, but if you are |
| 114 |
interested, row i corresponds to an EDSS score of i/2 (e.g. the fifth |
| 115 |
row is for EDSS=2.5) and column j corresponds to a duration of j-1 |
| 116 |
years. |
| 117 |
|
| 118 |
Output files |
| 119 |
=========== |
| 120 |
|
| 121 |
MSSStest produces the two output files. |
| 122 |
|
| 123 |
msss.out |
| 124 |
This contains a summary of the results of the Kruskal-Wallis test or in |
| 125 |
the case of a single test just the mean MSSS. |
| 126 |
|
| 127 |
indivmsss.out |
| 128 |
This is the same as edss.txt, except that an extra, sixth column is |
| 129 |
added which contains each individual's MSSS scores. |
| 130 |
|
| 131 |
|
| 132 |
Obtaining MSSStest |
| 133 |
================== |
| 134 |
|
| 135 |
Versions of MSSStest are available for Windows and Linux - see |
| 136 |
the download page: |
| 137 |
|
| 138 |
http://www-gene.cimr.cam.ac.uk/MSgenetics/GAMES/MSSS/download.html |
| 139 |
|
| 140 |
If you wish to use the program on a platform other than Windows or |
| 141 |
Linux, we also provide source code on the download page. MSSStest is |
| 142 |
written in C++ and may be compiled using a suitable C++ compiler. The |
| 143 |
files roxburgh.cpp and neededroutines.cpp are C++ source files and |
| 144 |
neededroutines.h is a "header" file. |
| 145 |
|
| 146 |
In unix, compilation would normally be by: |
| 147 |
|
| 148 |
CC -o MSSStest.exe neededroutines.cpp roxburgh.cpp -lm -lc |
| 149 |
|
| 150 |
where CC is the command name of the compiler. This produces the |
| 151 |
executable file MSSStest.exe. Alternatively, a makefile is also |
| 152 |
included (this may require editing for your C++ compiler). |
| 153 |
|
| 154 |
Testing the program |
| 155 |
=================== |
| 156 |
|
| 157 |
We provide a test file, edss.dat. This contains 250 patients, |
| 158 |
belonging to three genotype groups. When these data are analysed using |
| 159 |
the Global MSSS, the p-value is 0.69. When they are analysed using |
| 160 |
Local MSSS (which we do not recommend for such a small dataset), it is |
| 161 |
0.62. |
| 162 |
|
| 163 |
The data files edss.txt and global.dat need to be in the same folder as |
| 164 |
the file mssstest.exe. The way the program is launched in Windows is |
| 165 |
simply by double clicking on the mssstest.exe icon (once it has been |
| 166 |
unzipped). |
| 167 |
|
| 168 |
|
| 169 |
Licensing of MSSStest |
| 170 |
===================== |
| 171 |
|
| 172 |
This software is downloaded at the user's risk. Anyone may copy it but |
| 173 |
no one may sell it. |
| 174 |
|
| 175 |
We make no representation or warranties with respect to the MSSStest |
| 176 |
software and specifically disclaim any implied warranties of |
| 177 |
merchantability and fitness for a particular purpose. We reserve the |
| 178 |
right to revise the MSSStest software and to make changes therein from |
| 179 |
time to time without obligation to notify any person or organisation of |
| 180 |
such revision or changes. While we make every effort to ensure the |
| 181 |
reliability of the MSSStest software we may not be held responsible for |
| 182 |
errors, omissions or other inaccuracies or any consequences thereof. |
| 183 |
|
| 184 |
THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED |
| 185 |
WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF |
| 186 |
MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN |
| 187 |
NO EVENT SHALL SHAUN SEAMAN OR RICHARD ROXBURGH OR THEIR EMPLOYERS BE |
| 188 |
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR |
| 189 |
CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF |
| 190 |
SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR |
| 191 |
BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, |
| 192 |
WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR |
| 193 |
OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF |
| 194 |
ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 195 |
|
| 196 |
Version 2.0 September 2004 |
| 197 |
========================== |
| 198 |
|
| 199 |
The new features of this version are |
| 200 |
A)The program will calculate mean EDSS where all patients are included |
| 201 |
in one group. This can then be used as a general descriptor of |
| 202 |
progression in that group. |
| 203 |
|
| 204 |
B)While the program still accepts data from patients with disease |
| 205 |
duration less than one year ie whose disease duration in is given as |
| 206 |
"0" and will calculate a nominal MSSS for these patients these patients |
| 207 |
are not included in the calculation of mean MSSS nor used in the |
| 208 |
Kruskal Wallis tests. Note entering all patients in a group with |
| 209 |
disease duration "0" will cause it to crash. |
| 210 |
|
| 211 |
Version 3.0: 1st August 2007 |
| 212 |
============================ |
| 213 |
|
| 214 |
Versions of the program prior to this date had an error which made the |
| 215 |
file indivmsss.out unreliable when patients of duration 0 were |
| 216 |
entered. Individuals' MSSS scores as listed in this file were |
| 217 |
correct. However they were assigned to the wrong group if there were |
| 218 |
patients with duration 0 in the input file. This did not affect the |
| 219 |
Kruskal Wallis analysis or p values but did mean that calculated mean |
| 220 |
scores for each group were unreliable. MSSStest version 3.0 has this |
| 221 |
bug corrected. We are grateful to Yoav Ben-Shlomo (Dept. of Social |
| 222 |
Medicine, University of Bristol) for bringing this error to our |
| 223 |
attention. |
| 224 |
|
| 225 |
Shaun Seaman and Richard Roxburgh |
| 226 |
2 August 2007 |
| 227 |
|
| 228 |
Max Planck Institute for Psychiatry |
| 229 |
Kraepelinstr. 2-10 |
| 230 |
Munich |
| 231 |
Germany |