Abstract
Background: Diagnosis of the myelodysplastic syndromes (MDS) is based on a bone marrow examination (BME), an invasive procedure. Many patients prefer to avoid the test, the diagnosis is missed or delayed, and effective treatment is denied. A noninvasive diagnostic procedure would allow the clinician to provide the appropriate care while avoiding the invasive procedure.
Patients and Methods: BME-established MDS patients were compared to age-matched patients with anemia in whom MDS was excluded by BME. A number of variables were tested. The chosen variables (gender, age, Hb, MCV, PLT, WBC) were combined with multivariate logistic regression, their optimal coefficients
(parameters) were determined, and a score (Y), reflecting the probability to have the disease, was calculated. Sensitivity and specificity of each predicted probability was calculated. The probability cut-offs for diagnosing or ruling out MDS were selected by the most suitable sensitivity and specificity. Positive and
negative predictive values (PPV, NPV) were calculated for the chosen cut-offs.
Results: Data from 48 MDS and 63 non-MDS patients were used to establish the model. WBC and MCV, were found to be significantly different between the two groups. For the regression model, the ROC curve was drawn with area under the curve (AUC) of 0.748 (95% CI, 0.656 to 0.84). We chose two cut-off values: Patients with Y ≥ 0.633 were found to have high likelihood for MDS, with a
PPV of 85%. Patients with Y ≤ 0.288 were found to have low likelihood of having MDS, with a NPV of 81%. We propose defining the first group as probable MDS (pMDS), and the second group as probably not MDS, (pnMDS). The model was validated with 40 additional patients who underwent BME (20 with and 20
without MDS).
Conclusions: Using easily available clinical and lab data, we can diagnose or exclude MDS in some patients with high confidence, avoiding BME. Future work will use larger cohorts of patients to improve and further validate the model.
Patients and Methods: BME-established MDS patients were compared to age-matched patients with anemia in whom MDS was excluded by BME. A number of variables were tested. The chosen variables (gender, age, Hb, MCV, PLT, WBC) were combined with multivariate logistic regression, their optimal coefficients
(parameters) were determined, and a score (Y), reflecting the probability to have the disease, was calculated. Sensitivity and specificity of each predicted probability was calculated. The probability cut-offs for diagnosing or ruling out MDS were selected by the most suitable sensitivity and specificity. Positive and
negative predictive values (PPV, NPV) were calculated for the chosen cut-offs.
Results: Data from 48 MDS and 63 non-MDS patients were used to establish the model. WBC and MCV, were found to be significantly different between the two groups. For the regression model, the ROC curve was drawn with area under the curve (AUC) of 0.748 (95% CI, 0.656 to 0.84). We chose two cut-off values: Patients with Y ≥ 0.633 were found to have high likelihood for MDS, with a
PPV of 85%. Patients with Y ≤ 0.288 were found to have low likelihood of having MDS, with a NPV of 81%. We propose defining the first group as probable MDS (pMDS), and the second group as probably not MDS, (pnMDS). The model was validated with 40 additional patients who underwent BME (20 with and 20
without MDS).
Conclusions: Using easily available clinical and lab data, we can diagnose or exclude MDS in some patients with high confidence, avoiding BME. Future work will use larger cohorts of patients to improve and further validate the model.
Original language | English GB |
---|---|
Pages (from-to) | S78-S78 |
Journal | Leukemia Research |
Volume | 55 |
Issue number | S1 |
DOIs | |
State | Published - Apr 2017 |