PREMM1,2 Model : Prediction model for mutation status of MLH1 and MSH2 genes
Judith Balmaña, MD; David H. Stockwell, MD, MPH; Ewout W. Steyerberg, PhD; Elena M. Stoffel, MD, MPH; Amie M. Deffenbaugh, BS; Julia E. Reid, MStat; Brian Ward, PhD; Thomas Scholl, PhD; Brant Hendrickson, MS; John Tazelaar, MD; Lynn Anne Burbridge, BS; Sapna Syngal, MD, MPH
The PREMM1,2 model is a clinical prediction rule designed to be used by healthcare professionals to estimate the probability that an individual carries a mutation in MLH1 or MSH2. Mutations in these genes are found in most patients with the Lynch syndrome. The model was derived from a logistic regression analysis performed in a development cohort and was subsequently prospectively validated in a validation cohort. The analysis was performed in a cohort of 898 consecutive, unrelated probands who submitted blood samples for full gene sequencing of MLH1 and MSH2 and supplied personal and family histories to Myriad Genetic Laboratories, Inc. The individuals tested were mainly from the United States. Variables included in the model related to the proband were the presence and age at diagnosis of colorectal cancer (CRC), colonic adenomas, endometrial cancer, and other Lynch syndrome-associated cancers (ovary, stomach, kidney/urinary tract, bile ducts, small bowel, brain tumors (glioblastoma multiforme), pancreas, and sebaceous gland tumors). Other Lynch syndrome-associated cancers were considered as one group. Variables related to the relatives included the number of relatives with CRC, endometrial cancer, and other Lynch syndrome-associated cancers, the relationship to the proband (1st versus 2nd degree), the minimum age at diagnosis for each cancer in the family, and the presence of a relative with more than one Lynch syndrome-associated cancer. Overall, 14.5 percent (130/898) of the probands carried a pathogenic mutation (MLH1=6.5 percent, MSH2= 8.0 percent). The PREMM1,2 model performed well in the development cohort. The area under the ROC (Receiver operating characteristics) curve was 0.81, and the Hosmer-Lemeshow goodness-of-fit statistic had a p value greater than 20. The model has been externally validated in a validation cohort of 1016 consecutive unrelated probands who were tested for full gene sequencing and large rearrangement analysis of MLH1 and MSH2 genes in the same laboratory. Most of the individuals tested were mainly from the US and provided their personal and family history in the same test order form used for the development cohort. Overall, 15.3 percent (155/1016) of the individuals carried a pathogenic mutation in the validation cohort, with 42 (27 percent) of the latter being large rearrangements. The final prediction model was based on logistic regression coefficients estimated from both cohorts. Strong predictors of mutations included proband characteristics (presence of colorectal cancer, especially two or more diagnoses, or endometrial cancer) and family history (especially the number of first-degree relatives with colorectal or endometrial cancer). Age at diagnosis was especially important for colorectal cancer. Validation of the model in the validation cohort showed a good discriminatory capacity with a ROC area of 0.80 (0.76-0.84). Sensitivity and specificity of the PREMM1,2 model depends on the cutoff used for predicted risk of mutation. If a low cutoff, such as 5 percent, is used, many patients would be considered for testing, with a sensitivity of 94 percent but a specificity of 29 percent. If a high cutoff, such as 40 percent, is used, specificity would be much better (92 percent), but many patients with mutations would be missed (sensitivity of 29 percent).
Health care professionals using this model should be aware of its limitations. The accuracy of the family history reported on the test form could not be verified, and information about certain diagnoses was limited. However, the fact that the test order form was completed by health care professionals likely minimizes reporting erroneous diagnosis. Secondly, data collection did not allow the authors to account for family size and unaffected individuals; rather the model is depending upon crude number of affected individuals. Finally, the model only provides estimation for MLH1 and MSH2 gene mutation status and updating of the model with MSH6 analysis will be provided when sufficient data are available.
More information is available in: Balmaña et al. Prediction of MLH1 and MSH2 mutations in Lynch Syndrome. JAMA, 2006.
