Development and validation of the predictive aplastic score system (PASS): A simplified tool to diagnose acquired aplastic anemia in adults Free

Acquired aplastic anemia (AA) is an immune-mediated bone marrow failure characterized by pancytopenia and a hypocellular marrow, requiring prompt treatment. However, AA shares overlapping features with inherited bone marrow failure syndromes (IBMFS), which require different management. Diagnosis relies on excluding IBMFS, but genetic testing, while helpful, is not always available, may delay care, and can yield inconclusive results. The goal of our study was to develop the Predictive Aplastic Score System (PASS), a simple, clinically applicable tool using readily available patient data to accurately distinguish AA from IBMFS.

The training cohort included 212 consecutive adult patients aged ≥18 years who were evaluated for suspected AA and bone marrow failure between 2010 and 2025. Of these, 162 (76.4%) were diagnosed with AA, while 50 (23.6%) were diagnosed with IBMFS using standard criteria, including genetic and functional testing. Among the IBMFS group, 25 (50.0%) had telomere biology disorders, 8 (16.0%) had Fanconi anemia, 5 (10.0%) had Diamond-Blackfan anemia, 3 (6.0%) had GATA2 deficiency, and 9 (18%) were diagnosed with other IBMFS.

AA patients presented at an older age (median 54.7 years; range 19.3–86.7) compared to IBMFS patients (median 37.4 years; range 18.8–72.4; p=0.001). A substantial proportion of patients in both groups were ≥60 years (AA 38.9% vs IBMFS 22.0%; p=0.029). Among 162 AA patients, 143 (88.3%) had acute-onset cytopenias of <1 year duration, compared to 8 of 50 (16.0%) IBMFS cases (p<0.001). 79.6% (129 of 162) of AA patients had severe or very severe AA (SAA/VSAA) as compared 8% (4 of 50) IBMFS patients (p<0.001). Lymphocyte telomere length (TL) <1st percentile occurred in 2.6% (2 of 77) AA patients compared to 56.4% (22 of 39) of IBMFS patients (p<0.001).

Using logistic regression with the least absolute shrinkage and selection operator (LASSO) method with 10-fold cross-validation, we selected seven clinical variables associated with the diagnosis of AA or IBMFS for inclusion in the PASS model: severity, acuity, age, IBMFS red flags, AA-associated conditions, AA-associated somatic changes, and lymphocyte TL<1^st percentile. To create the PASS score, we assigned points based on LASSO coefficient and refined by expert consensus, with positive values indicating acquired AA and negative values indicating IBMFS.

In the training cohort, the final model achieved an area under the Receiver Operating Characteristic (ROC) curve (AUC) of 0.991 (95% Confidence interval (CI) 0.983–0.999). Higher score thresholds were highly specific for AA, with scores ≥30 capturing only AA patients and no IBMFS patients, allowing for rapid classification of 86.4 % (140 of 162) of AA cases in the training cohort. Among patients with intermediate scores (0 to 20), 79.2% (19 of 24) had AA. In contrast, the majority (86.5%. 45 of 52) of patients with scores <0 had IBMFS.

Validation of the PASS score in three independent external bone marrow failure cohorts of 78, 121, and 247 patients each demonstrated excellent discriminatory performance, with ROC AUC of 0.914 (95% CI: 0.783–1.000), 0.982 (CI 0.964–0.999), and 0.969 (CI 0.948–0.989), respectively. To test inter-rater reliability, three independent raters evaluated a sample of 20 patients, with excellent inter-rater agreement of 0.766 (95% CI 0.586–0.883). A threshold-based analysis on the three combined cohorts, including 344 AA and 103 IBMFS patients, confirmed a PPV of 100% for the diagnosis of AA in patients with scores ≥30, allowing for rapid diagnosis of AA in 65.9% of AA patients.

In conclusion, we developed a practical, accurate, and readily deployable clinical scoring tool that can rapidly distinguish acquired AA from IBMFS at diagnosis for over two-thirds of AA patients. This approach is more efficient and cost-effective by minimizing diagnostic delays, reducing healthcare utilization, and enabling earlier treatment of AA, while identifying patients at higher risk of IBMFS for whom we should prioritize genetic testing. The strong performance of the PASS score in both the training and validation cohorts supports its application in diverse clinical settings, including those with limited access to advanced diagnostics, and can guide timely and appropriate management. To promote clinical adoption, we developed an open-access web-based calculator.