Fig 1.
Specimen work-up and heterogeneous spatial distribution of the number of follicles on one slide A) Work-up scheme for the specimens in the MGTX trial [1, 3]. Thymuses were sub-divided into regions that underwent predefined evaluation (e.g. complete work-up of the central A-region, partial work-up of other regions) B) Distribution of the median number of lymphoid follicles in CD23-stained sections per case and the associated minimum-maximum-value-range across all available slides per case. Case 37T004 from the MGTX-trial is highlighted as an example showing high in-case variation.
Table 1.
Overview of gathered histomorphological data.
For most of the slides of the 55 thymectomy specimens 16 histomorphological parameters were collected using different means of measurement as indicated in column. Region “A” designates the completely processed central horizontal tissue plane of a given thymus as shown in Fig 1A. HE, hematoxylin and eosin.
Fig 2.
Variance of the entropy in the MGTX collective.
Entropy was calculated for the number [n] of follicles and for the grading of lymphofollicular hyperplasia in CD23-stained sections [grades 0–4]; entropy calculations for the grading of intra-thymic fat content [%] and the grading of thymic atrophy [grades 0–4] were based on measurements based on HE-stained sections. Diagonal: The histogram per entropy is plotted with case 37T004 being highlighted. Lower triangle: Below the diagonal the variables are plotted against each other. Upper triangle: Above the diagonal the corresponding correlation coefficients are plotted.
Fig 3.
Correlation between the entropy of intra-thymic fat and post-operative absolute prednisone-dose.
On linear modelling, a significant correlation (p = 0.03) between the entropy of intra-thymic fat per case and the post-operative absolute prednisone dose (area under the dose-time curve for prednisone months 0–12) could be shown. To visualize the correlation the values per case are plotted against each other with the calculated regression line.
Table 2.
To model the effect of the entropy on certain clinical parameters / endpoints, explorative modelling with logistic regression for categorical clinical endpoints (QMG-drop, MMS) and linear regression for clinical endpoints with linear scaling (QMG-score, MG-duration, pre- and postoperative prednisone load) were performed. Respectively, one clinical parameter is thereby modeled on basis of four entropy values (for the number of follicles in the CD23-staining, the follicle grading in the CD23-staining, the grading of the intratyhmic fat and the grading of the atrophy). The table shows the p-value (* indicates significance at the 5%-level) per variable in the adjusted model that contains all the other variables against the respective clinical endpoint.
Fig 4.
Scheme to illustrate the definition of levels and the calculation of entropy that is based thereon.
A-B): For one fictional case the spatial heterogeneity and the in-case variance for one variable is illustrated (number of follicles per slide in Fig 2A and atrophy grading in Fig 2B). The numbers of follicles per slides were mapped to one of 10 levels (with an interval of 5 lymphoid follicles; accordingly level 0 = <5 follicles; level 1 = 5 - < 10; etc.). The resulting frequencies / relative probabilities are plotted next to the scheme. C-D): Entropy as calculated for the number of follicles in case 37T004 C) Fictional thymus with a non-random, completely homogenous distribution of measurements for the number of follicles. The corresponding entropy equals zero. D) Mean-values and standard deviations of the number of follicles per region in CD23-stained sections of case 37T004. The entropy value in this case with heterogeneously distributed follicle frequencies across the various regions is 2.2bit.