PET-Based Metabolic Score Outperforms Standard Risk Model in Lymphoma Patients

Diffuse large B-cell lymphoma, the most common form of non-Hodgkin lymphoma worldwide, has long been staged and risk-stratified using clinical tools that were developed decades before modern imaging became routine. Now, a retrospective real-world study from China suggests that a metabolic scoring system built on positron emission tomography data can sharpen the risk picture for a large group of patients whose outlook is otherwise difficult to predict. The research, published in BMC Cancer, compares the International Metabolic Prognostic Index, known as IMPI, against the widely used National Comprehensive Cancer Network version of the International Prognostic Index, or NCCN-IPI, in patients receiving first-line treatment with the immunochemotherapy regimen R-CHOP.

The clinical stakes are considerable. The NCCN-IPI, a refinement of the classic International Prognostic Index, assigns patients to risk categories using factors such as age, lactate dehydrogenase levels, disease stage, extranodal involvement and performance status. Yet within any single NCCN-IPI category, outcomes vary widely: some patients deemed low or intermediate risk relapse early, while others classified as high risk do remarkably well. Clinicians have therefore searched for biological markers that capture the true aggressiveness of the disease rather than relying solely on anatomical spread and laboratory values. Metabolic tumour volume, measured from fluorodeoxyglucose PET scans, has emerged as one of the most promising candidates because it quantifies the total burden of actively metabolising tumour tissue in a single number.

The IMPI was conceived to exploit exactly that information. Instead of counting nodal sites or anatomical stages, the index incorporates total metabolic tumour volume, or TMTV, alongside other prognostic variables to sort patients into risk groups. Earlier work derived the original IMPI from clinical trial populations, which tend to be younger, fitter and more carefully selected than the patients who show up in everyday oncology clinics. That raised an obvious question: does the index still work when applied to a real-world cohort? The new study set out to answer it, and also to address a technical gap. The original IMPI model did not incorporate lactate dehydrogenase, which made direct comparison with the NCCN-IPI, a model that does include LDH, statistically awkward.

To conduct the comparison, the researchers assembled a cohort of 294 patients with newly diagnosed diffuse large B-cell lymphoma who had all been treated with first-line R-CHOP, the standard combination of rituximab, cyclophosphamide, doxorubicin, vincristine and prednisone. Each patient underwent baseline fluorodeoxyglucose PET/CT imaging, from which the team calculated total metabolic tumour volume using a standardised uptake value threshold of 4 and, in a parallel exploratory analysis, a semi-automatic threshold set at 41 percent of the maximum SUV within each lesion, denoted TMTV41%. Patients were then classified both into four categorical IMPI subgroups and into three continuous IMPI risk strata, allowing the investigators to test whether the index performed better as a stepped scale or as a graded variable.

The central finding concerns the patients that conventional scoring leaves in a grey zone. Within the NCCN-IPI low-risk, low-intermediate-risk and high-intermediate-risk populations, patients who were upgraded to a higher risk category by the IMPI consistently showed worse outcomes across three endpoints: time to progression, progression-free survival and overall survival. In other words, when the metabolic index disagreed with the clinical index and painted a darker picture, the metabolic index tended to be right. The signal was not perfectly uniform, however. Individual comparisons for time to progression in the low-risk group and overall survival in the low-intermediate-risk group did not reach statistical significance, a reminder that even strong prognostic trends can wobble in subgroups of modest size.

Equally telling was what happened on the other side of the ledger. Among patients classified as high risk by the NCCN-IPI but downgraded to a lower category by the IMPI, no survival improvement was observed. This asymmetry matters for clinical interpretation. It suggests that the metabolic index is most useful for identifying hidden risk within apparently favourable groups, not for reassuring patients whose clinical scores look ominous. A patient flagged as high risk by the NCCN-IPI should probably not be downgraded on the basis of a favourable metabolic volume alone, at least not on the strength of the current evidence.

To move beyond eyeballing survival curves, the researchers quantified model discrimination using two complementary statistics. Harrell’s concordance index, or c-index, measures how well a model ranks patients relative to one another, with values closer to 1 indicating better discrimination. The Akaike Information Criterion, or AIC, rewards models that fit the data well while penalising unnecessary complexity, with lower values indicating a better balance. For time to progression and progression-free survival, both the categorical and continuous versions of the IMPI posted numerically higher c-index values and lower AIC values than the competing models. For overall survival, only the categorical and continuous exploratory IMPI41% variants showed preferable numerical indicators, and the continuous IMPI41% delivered the best numerical performance across all models tested.

The technical details behind those measurements deserve attention, because they speak to the reproducibility of metabolic imaging biomarkers. Total metabolic tumour volume is sensitive to the threshold used to delineate tumour from background uptake. Fixed SUV thresholds, such as the SUV of 4 used in the primary analysis, are simple but can behave differently across scanners and patient body compositions. Relative thresholds, such as the 41 percent of SUVmax cutoff applied in the exploratory analysis, adapt to each lesion’s peak activity and have been endorsed in guidance from the European Association of Nuclear Medicine for volumetric lymphoma assessment. The fact that the exploratory IMPI41% variant performed best numerically in this cohort adds to a growing body of evidence that relative thresholding may capture tumour burden more faithfully, though the authors are careful to frame it as exploratory rather than definitive.

What does this mean for patients and their physicians? The most immediate implication is that a substantial fraction of DLBCL patients, those sitting in the low, low-intermediate and high-intermediate NCCN-IPI categories, could in principle be subdivided more accurately using information already contained in their baseline PET scans. If validated externally, IMPI-based stratification could help identify patients who might warrant closer surveillance, earlier response assessment, or enrollment in trials of intensified or novel therapies. The authors are explicit, however, that the study does not establish therapeutic benefit. It cannot be confirmed from these data that the high-risk subgroups screened out by the IMPI would actually gain from early intensive treatment, and changing therapy on the basis of this index alone would be premature.

The study also carries the usual caveats of retrospective, single-region research. All 294 patients were treated at centres affiliated with Jiangsu Cancer Hospital and Jiangsu Provincial Cancer Hospital in Nanjing, and the protocol was approved by the hospital’s ethics committee with the informed consent requirement waived because of the retrospective design. Cohort size limits the precision of subgroup analyses, and the survival differences that failed to reach significance in some comparisons may simply reflect small numbers rather than true absence of effect. The authors themselves call for external validation in larger, independent cohorts before the findings are translated into practice. Still, the direction of the evidence is coherent: metabolic tumour burden, properly thresholded and properly modelled, appears to carry prognostic information that the clinical indices miss, and the IMPI, particularly in its continuous IMPI41% form, may be the most efficient way yet devised to extract it from images oncologists already acquire.

Subject of Research: Prognostic risk stratification of diffuse large B-cell lymphoma using metabolic tumour volume-based indices compared with the NCCN-IPI in R-CHOP-treated patients

Article Title: IMPI refines prognostic stratification among non-high NCCN-IPI risk DLBCL patients treated with R-CHOP: a real-world cohort study

Article References: Cui, S., Jiang, Q., Zhang, H., Wen, Z., Luan, P., Hu, Y., & Guo, S. (2026). IMPI refines prognostic stratification among non-high NCCN-IPI risk DLBCL patients treated with R-CHOP: a real-world cohort study. BMC Cancer. https://doi.org/10.1186/s12885-026-16926-y

Image Credits: AI Generated

DOI: 10.1186/s12885-026-16926-y

Keywords: diffuse large B-cell lymphoma, IMPI, NCCN-IPI, total metabolic tumour volume, PET/CT, R-CHOP, prognosis, risk stratification, progression-free survival, overall survival, real-world cohort, hematologic malignancy