The index was calculated from fasting triglyceride, high-density lipoprotein cholesterol and glucose concentrations. Researchers also used receiver operating characteristic curve analysis to identify the TyHGB value that best classified participants as having high ASCVD risk. That cutoff was 8.44, but its discriminatory performance was modest: the area under the curve was 0.607, with a 95% confidence interval of 0.540–0.673. The result indicates limited ability to distinguish high-risk from lower-risk participants when TyHGB is used alone, underscoring that the observed association does not amount to strong standalone prediction.

With TyHGB treated as a continuous variable, the fully adjusted odds ratio for high ASCVD risk was 1.21 per unit increase, with a 95% confidence interval of 1.08–1.35 and P<0.001. When the index was divided at the ROC-derived cutoff, the high-versus-low comparison produced an odds ratio of 4.02, with a 95% confidence interval of 2.13–7.56 and P<0.001. These estimates describe odds within this study population at a single assessment; they do not establish the direction of causality, changes over time, or the effect of changing TyHGB. The supplied results report the significant association for the fully adjusted model rather than describing it as independently present in every sequential adjustment model.

Restricted cubic spline analysis suggested a linear dose-response relationship between TyHGB and high ASCVD risk. The test for nonlinearity returned P=0.053, meaning the study did not detect statistically significant departure from a linear pattern at the reported threshold. The result is consistent with progressively greater odds of high-risk classification across higher index values, although the cross-sectional design cannot show whether increasing the index would be followed by a change in cardiovascular risk. It also does not establish that TyHGB is a useful measure for monitoring future risk or for guiding clinical decisions.

Subgroup analyses indicated that the strength of the association varied according to two clinical characteristics. Interaction tests were significant for alcohol use, with P=0.024, and for hypertension, with P=0.041, indicating effect modification by these factors in the study analyses. The source does not establish whether TyHGB would be useful for screening or clinical decision-making, or whether the relationship applies beyond similar populations. The study’s principal contribution is the observed association in a retrospective, single-center cross-sectional analysis of adults with NAFLD, combined with evidence that the index alone distinguishes ASCVD risk categories only weakly. Because the design assesses participants at one point in time and the reported association is observational, the findings should not be interpreted as proof that changing the index would change cardiovascular outcomes. Larger prospective studies would be needed to assess whether the relationship persists over time, applies in broader populations, or provides useful information beyond other cardiovascular risk measures.

Subject of Research: Association between the TyHGB index and high atherosclerotic cardiovascular disease risk in adults with non-alcoholic fatty liver disease

Article Title: Association between the TyHGB index and the high risk of ASCVD in Chinese patients with NAFLD: a cross‑sectional study

Article References: Xiong, L., & Zhuang, H. (2026). Association between the TyHGB index and the high risk of ASCVD in Chinese patients with NAFLD: a cross‑sectional study. BMC Endocrine Disorders. https://doi.org/10.1186/s12902-026-02678-0

Image Credits: AI Generated

DOI: 10.1186/s12902-026-02678-0

Keywords: TyHGB index, NAFLD, ASCVD risk, cardiovascular disease, triglycerides, glucose, cross-sectional study, odds ratio, subgroup analysis, Association, between, TyHGB

The research team, led by Nu Hoang and Thin Nguyen of Deakin University’s Applied Artificial Intelligence Institute, together with collaborators from the university’s Global Centre for Preventive Health and Nutrition, set out to solve a problem that has long undermined evaluations of community interventions. Traditional statistical approaches, such as linear regression, demand careful selection of covariates to ensure valid causal inference. Choosing the wrong variables, either omitting a genuine confounder or adjusting for an inappropriate one, can bias estimates of a program’s effect. Even with the right covariates, misspecifying the functional form of the model can still produce misleading conclusions. The team’s answer was not a new algorithm but a principled, end-to-end workflow that chains together causal discovery, formal identification through do-calculus, double machine learning, and interpretable subgroup discovery.

The first stage of the pipeline is causal discovery, the task of inferring cause-and-effect relationships directly from data rather than assuming them from prior theory. The researchers employed a hybrid causal discovery model designed for mixed-type data, meaning it can handle both continuous measurements, such as body mass index, and categorical variables, such as sex. The model works in two phases: a randomized conditional independence test builds the skeleton of the causal graph, and a cross-validation-based scoring function then orients and refines the edges. Because community intervention data are inherently longitudinal, the team adapted the model with temporal constraints, learning a graph over baseline variables first and then forcing edges to flow forward in time. This two-step design proved its worth in validation: it achieved significantly higher log-likelihood on held-out data than a one-step alternative, with a p-value below 0.0001.

The resulting causal graph is a map of how children’s characteristics and behaviors interconnect. Baseline age stood out as the most influential upstream variable, with the highest mean out-degree of 13.4 across bootstrap resamples, linking to physical activity, sedentary behavior, active transport, fruit intake, sweet beverage consumption, takeaway food consumption, and the outcome itself: the change in body mass index z-score over time. Baseline zBMI and sex were the next most recurrent upstream variables. Critically, the confounding pathways from sex, baseline age, and baseline zBMI to program participation appeared with a bootstrap frequency of 1.0, meaning they were perfectly stable across repeated resampling of the data. This stability gave the researchers confidence that the adjustment set identified from the graph could reliably strip away spurious correlations.

With the causal structure established, the team turned to double machine learning, a technique from econometrics that allows flexible machine learning models to be used in causal estimation without sacrificing statistical consistency. The method models both the outcome and the treatment assignment as functions of the covariates, then exploits two tricks to avoid the biases that normally plague machine learning estimators. Cross-fitting splits the data into partitions so that models are never evaluated on the data used to train them, guarding against overfitting bias. Orthogonalization, rooted in the classical Frisch-Waugh-Lovell theorem, regresses out nuisance functions and estimates the causal parameter from residuals, delivering a root-N consistent estimator free of regularization bias. Here, LightGBM models served as the nuisance learners, with hyperparameters tuned by fivefold cross-validated grid search.

The headline result of the estimation stage is a picture of profound heterogeneity. The average treatment effect across all children was a modest reduction of 0.02 in zBMI, with a 95 percent confidence interval spanning from minus 0.09 to plus 0.05, a range that crosses zero. But the individualized estimates told a far richer story, ranging from minus 0.25 to plus 0.15. Roughly 18.7 percent of children had confidence intervals entirely below zero, indicating a genuine benefit, while 8.7 percent had intervals entirely above zero, suggesting the program may have been counterproductive for them. The remaining 72.6 percent had intervals that included zero, reflecting the substantial uncertainty inherent in individual-level causal estimates. The researchers are careful to stress that these individualized figures describe broad patterns of heterogeneity rather than decision-grade predictions for any single child.

To probe what drives this variation, the team first examined linear correlations between the estimated effects and baseline variables. Age showed the strongest association, with a Pearson coefficient of 0.59, while no other baseline variable correlated significantly. Because linear correlation can miss nonlinear structure, the researchers then trained a shallow decision tree to classify children into positive-effect, negative-effect, and no-effect groups based on nine baseline characteristics. The tree partitioned the data using just two variables: age and weekly takeaway food consumption. Children under 8.9 years old formed the most robustly benefited subgroup, with a dominant-class probability of 0.93 and a mean individual treatment effect of minus 0.096, more than four times the pooled average. Children aged 8.9 to 13.6 who ate takeaway food less than once per week also benefited, with a mean effect of minus 0.038. In contrast, adolescents above 15.2 years showed a clearly negative response, with a mean effect of plus 0.018.

The robustness of these findings was tested from multiple angles. Bootstrap validation of the decision tree across 100 resampled datasets confirmed that age was selected as a splitting variable in every single run, while takeaway food consumption appeared in 40 percent of runs and no other variable appeared at all. Refutation tests bolstered the causal estimate itself: a placebo treatment test, which replaces the real treatment with random values, yielded an effect of minus 0.0038 that was statistically indistinguishable from zero, exactly as expected if the original estimate reflects a genuine causal relationship. Subset validation produced an effect of minus 0.016, not significantly different from the original. An E-value sensitivity analysis indicated that an unmeasured confounder would need to be associated with both treatment and outcome by a risk ratio of at least 1.23 to explain away the average effect, though the subgroup effect for younger children is substantially larger and more resilient.

The authors are candid about the limitations of their work. Communities self-selected into the intervention programs, so unmeasured factors such as local political support or socioeconomic resources could confound the results, potentially overstating benefits. Spillover effects, in which children in comparison communities are indirectly exposed to intervention activities, could bias estimates toward the null. The pooled average effect is small and its confidence interval crosses zero, so the researchers emphasize that the study’s main contribution lies in demonstrating how heterogeneity can be identified and characterized, not in claiming a large overall benefit. The individualized estimates are explicitly framed as exploratory and potentially model-dependent, useful for revealing broad subgroup structure rather than for clinical decision-making at the level of a single child.

Even with those caveats, the implications are considerable. If community-based obesity prevention is genuinely most effective for younger children and least effective for older adolescents, then policymakers have a concrete, actionable signal: interventions may need age-sensitive redesign, with different strategies for teenagers than for primary-school children. The interaction analysis adds nuance, finding that the combination of age and takeaway food consumption drives heterogeneity beyond age alone, and that physically active children who eat fewer takeaway meals benefit more. Beyond obesity, the framework is explicitly generalizable. The authors argue that the same workflow, causal discovery to build the graph, do-calculus to identify the adjustment set, double machine learning to estimate effects, and tree-based subgroup analysis to interpret them, could be applied to any complex community intervention where average effects conceal the individuals the program actually helps. In an era when public health budgets are finite and one-size-fits-all programs increasingly look inadequate, that may be the study’s most viral-worthy message: the average is a lie, and the tools to see past it now exist.

Subject of Research: Causal machine learning analysis of heterogeneous treatment effects in community-based childhood obesity prevention interventions

Article Title: Causal machine learning for understanding heterogeneous effects of childhood obesity prevention

Article References: Hoang, N., Nguyen, T., Duong, B., Nichols, M., Brown, V., Backholer, K., Allender, S., & Nguyen, T. (2026). Causal machine learning for understanding heterogeneous effects of childhood obesity prevention. International Journal of Data Science and Analytics, 22(1), Article 326. https://doi.org/10.1007/s41060-026-01156-z

Image Credits: AI Generated

DOI: 10.1007/s41060-026-01156-z

Keywords: causal machine learning, causal discovery, causal inference, childhood obesity, community-based interventions, heterogeneous treatment effects, double machine learning, BMI z-score, public health, age heterogeneity, subgroup analysis, preventive health

Attention-deficit/hyperactivity disorder, commonly abbreviated as ADHD, is characterized by persistent patterns of inattention, hyperactivity, and impulsivity that interfere with functioning at school and at home. While stimulant and non-stimulant medications remain the best-established treatments, many parents and clinicians are eager for non-pharmacological alternatives, whether because of side effects, incomplete medication response, or family preference. Computerized cognitive training has emerged as one of the most heavily marketed options in this space, promising to strengthen the underlying attentional and executive systems through repeated, game-like practice. The scientific evidence, however, has been inconsistent, with some trials reporting gains on trained tasks and symptom scales and others finding little more than placebo-like effects.

The intervention tested in the new trial, known as IBT-AC01, is a multicomponent cognitive training program developed with input from medical advisors, including some of the study’s authors, and technical support from Infinite Brain Technologies. Rather than targeting a single skill such as working memory, the program combines multiple training tasks aimed at different executive functions, the family of cognitive control processes that includes sustained attention, inhibitory control, and cognitive flexibility. The rationale is that ADHD is a heterogeneous condition and that a broader training battery might engage a wider network of neural systems than single-task programs, which have often shown benefits that fail to generalize beyond the exercises themselves.

To test the program fairly, the researchers conducted a double-blind, parallel-group randomized controlled trial involving 105 children aged 6 to 12 years who had been diagnosed with ADHD. Participants were randomly assigned in equal numbers to receive either the IBT-AC01 training or a control condition. Children in the training arm completed daily sessions lasting 25 minutes, five days a week, for a total of four weeks, a dosing schedule designed to deliver an intensive but manageable course of practice. The trial was registered with the Chinese Clinical Trial Registry under number ChiCTR2200057306 before recruitment, and it was approved by the Ethics Committee of Shenzhen Children’s Hospital in accordance with the Declaration of Helsinki, with written informed consent obtained from all parents or guardians and from the children themselves when they were eight years old or older.

The investigators chose two primary outcomes to capture both laboratory-measured cognition and real-world clinical symptoms. The first was the Attention Comparison Score on the Test of Variables of Attention, or TOVA, a computerized continuous performance test widely used to quantify sustained attention and impulsivity. The second was the total score on the ADHD Rating Scale-IV, a clinician- and parent-oriented instrument that tallies the severity of the eighteen diagnostic symptoms of ADHD. Statistical models adjusted for baseline scores, age, and sex were used to compare the two groups, and the analysis followed an intention-to-treat approach, meaning that every randomized participant was analyzed according to their original assignment regardless of adherence, a conservative strategy that preserves the randomization and reduces bias.

The headline result was sobering for proponents of cognitive training. In the primary analysis, there were no statistically significant differences between the training group and the control group on either primary outcome. For the TOVA Attention Comparison Score, the adjusted between-group difference was a modest 0.33 points, with a 95 percent confidence interval spanning from minus 0.54 to 1.20 and a p-value of 0.45, well short of conventional significance thresholds. For the ADHD Rating Scale-IV total score, the adjusted difference was minus 1.07 points, with a confidence interval from minus 3.73 to 1.59 and a p-value of 0.43. In plain terms, after four weeks of intensive daily training, the children who completed the multicomponent program could not be distinguished from controls on either objective attention testing or rated symptom severity.

Yet the trial did not stop at the overall null result. In a methodologically interesting twist, the team applied an exploratory technique called model-based recursive partitioning, a form of statistical decision-tree analysis that searches a data set for subgroups of participants who respond differently to treatment. Rather than pre-selecting candidate subgroups based on theory alone, this approach lets the data reveal where treatment effects may diverge. The analysis flagged three characteristics associated with potentially greater benefit from IBT-AC01: younger age, the presence of comorbid oppositional defiant disorder, and higher baseline ADHD symptom severity. Children with more severe starting symptoms and those whose ADHD was complicated by defiant and oppositional behavior appeared, in this exploratory analysis, to derive more advantage from the training than the sample as a whole.

These subgroup findings come with important caveats that the authors themselves emphasize. Exploratory analyses of this kind are hypothesis-generating rather than confirmatory; because the subgroups were identified after the fact within a modest sample of 105 children, the signals could reflect statistical noise, and the confidence in any single partition is limited. Still, the patterns are biologically and clinically plausible. Younger children may have more malleable attentional systems and greater room for developmental gains, while children with more severe baseline symptoms have more distance to travel, leaving more scope for measurable improvement. The link with comorbid oppositional defiant disorder is intriguing and suggests that cognitive training might interact with the broader behavioral profile of a child in ways that uniform group averages conceal.

The trial’s design strengths deserve attention in a field where commercial cognitive-training products have often been evaluated with weak methodology. The double-blind structure, active control condition, registered protocol, intention-to-treat analysis, and adjustment for baseline covariates all reduce the risk of inflated effects. At the same time, the study raises questions about dosing and duration: four weeks of training, while intensive, may simply be too short to produce durable changes in the neural circuits supporting attention, or the gains may be too narrow to register on generalized outcome measures. Prior research on cognitive training has repeatedly documented the gap between improvement on trained tasks and transfer to everyday functioning, a phenomenon sometimes called the transfer problem, and the present results are consistent with that literature.

For families and clinicians, the practical message is one of tempered expectations. Multicomponent computerized cognitive training, at least in the form and dose tested here, cannot currently be recommended as a stand-alone treatment for ADHD in children, and the overall evidence base continues to suggest that its effects, when present, are modest and uneven. The exploratory subgroup results, however, offer a constructive direction for future research: larger trials designed from the outset to test whether younger children, those with comorbid behavioral difficulties, or those with severe symptoms respond preferentially could clarify who, if anyone, benefits most. The authors call for studies with larger samples to confirm efficacy and to probe the mechanisms underlying any differential response. Until then, the trial stands as a reminder that in cognitive intervention research, the average effect can hide as much as it reveals, and that careful, registered, randomized evidence remains the only reliable compass for navigating a crowded and commercially appealing treatment landscape.

Subject of Research: Efficacy of multicomponent computerized cognitive training for children with ADHD in a randomized controlled trial

Article Title: Efficacy of multiple-component cognitive training for ADHD children: a randomized controlled trial with subgroup analysis

Article References: Wu, Y., Xu, L., Cao, X., Wu, Z., Wang, Y., Xue, G., & Yang, B. (2026). Efficacy of multiple-component cognitive training for ADHD children: a randomized controlled trial with subgroup analysis. BMC Psychiatry. https://doi.org/10.1186/s12888-026-08683-w

Image Credits: AI Generated

DOI: 10.1186/s12888-026-08683-w

Keywords: ADHD, cognitive training, randomized controlled trial, children, executive function, TOVA, ADHD Rating Scale-IV, oppositional defiant disorder, non-pharmacological intervention, subgroup analysis, BMC Psychiatry, Shenzhen Children's Hospital

The original study, led by Lindner and colleagues, analysed tumours from 193 treatment-naïve patients with stage II melanoma, a disease stage in which the primary tumour is thick but has not yet spread to distant sites, and in which clinicians urgently need better tools to decide who requires intensive surveillance or adjuvant therapy. Among the many alterations catalogued in that work, one finding stood out: a deletion spanning the chromosomal region 11q23.1-3, which contains the CBL gene, was associated with relapse-free survival in the overall cohort when tested in a rigorous multivariate analysis. The Dalian-based team is quick to acknowledge the strength of that result, describing it as a compelling foundation for exploring CBL loss as a potential prognostic biomarker. Their concerns arise not from the headline finding, but from what happens when that finding is pushed into finer molecular subgroups.

CBL is no incidental passenger. The gene encodes an E3 ubiquitin ligase, a molecular machine that tags proteins for degradation and thereby acts as a critical regulator of signalling pathways driven by receptor tyrosine kinases. Because melanomas are classically stratified by driver mutations in BRAF, RAS and NF1, with a residual group classified as triple wild-type, any genomic alteration that appears to carry prognostic weight within one of these subtypes immediately attracts attention. The Lindner study reported that the 11q23.1-3 deletion showed a prognostic trend within the RAS-mutated subgroup of patients, and it is precisely this subgroup-specific claim that Han, Gao and Jiang dissect in their correspondence.

The heart of their critique concerns the difference between an unadjusted p-value and an adjusted one. In the RAS-mutated subgroup, the original report highlighted an unadjusted p-value of 0.044 for relapse-free survival, a figure that sits just below the conventional 0.05 threshold and therefore appears, at first glance, to signal genuine statistical significance. But when the analysis was corrected for multiple testing, the adjusted p-value rose to 0.178, well above the threshold that most researchers would accept as evidence of a reliable effect. The distinction is far from pedantic. When investigators test many genomic subgroups simultaneously, as happens when BRAF, RAS, NF1 and triple wild-type tumours are each interrogated for prognostic associations, the probability of stumbling across at least one apparently significant result by pure chance rises steeply. This phenomenon, known as a Type I error, is the false positive that multiple-testing adjustments are designed to suppress.

Han and colleagues argue that in exploratory subgroup analyses spanning multiple genomic subtypes, adjusting for multiple comparisons is generally recommended to prevent exactly these spurious discoveries. They point to the influential 2007 New England Journal of Medicine commentary by Wang, Lagakos, Ware, Hunter and Drazen on the reporting of subgroup analyses in clinical trials, a paper that has shaped how statisticians and clinicians interpret claims carved out of broader datasets. That commentary warned that subgroup findings are frequently overinterpreted, particularly when unadjusted significance levels are emphasized over corrected ones. By foregrounding the unadjusted p-value of 0.044 while the adjusted figure of 0.178 tells a more cautious story, the original presentation, the correspondents suggest, risks conveying a degree of predictive confidence in the RAS-mutated subgroup that the data do not yet support.

There is also the matter of sub-stage confounding, a second analytical nuance the letter raises. Stage II melanoma is not a single homogeneous category. Under the American Joint Committee on Cancer eighth edition staging system, refined in the landmark 2017 update by Gershenwald and colleagues, stage II encompasses patients with tumours of markedly different thicknesses and ulceration statuses, and these features themselves carry powerful prognostic information. When a cohort is subdivided first by molecular subtype and then examined for survival associations, imbalances in tumour thickness, ulceration or other clinicopathological variables between patients with and without the CBL region deletion can masquerade as genuine biological effects. Disentangling whether the deletion independently forecasts relapse, or merely travels alongside known risk factors that happen to cluster within the subgroup, demands careful covariate adjustment and transparent reporting of how residual confounding was handled.

The correspondents anchor their argument in established reporting standards, invoking the REMARK guidelines, the Reporting Recommendations for Tumor Marker Prognostic Studies published by McShane and colleagues in 2005 in the Journal of the National Cancer Institute. REMARK was developed precisely because biomarker prognostic studies have historically been plagued by small samples, selective reporting and optimistic interpretation, leading to markers that fail repeatedly upon validation. The guidelines call for complete documentation of statistical methods, prespecified hypotheses, transparent handling of multiple testing and honest characterisation of exploratory versus confirmatory findings. Emphasising adjusted p-values, Han, Gao and Jiang contend, aligns with these norms and helps readers accurately gauge the robustness of the CBL alteration within specific molecular subsets rather than being swept up in an apparently significant number.

None of this diminishes the value of the underlying discovery. The genomic classification of cutaneous melanoma established by The Cancer Genome Atlas Network in 2015 demonstrated that melanoma biology divides cleanly into the BRAF-mutant, RAS-mutant, NF1-mutant and triple wild-type categories, and subsequent efforts to layer prognostic information onto that framework have been a major research priority. A driver gene and biomarker candidate emerging from a 193-patient cohort of therapy-naïve stage II patients is genuinely noteworthy, particularly for a disease stage in which sentinel lymph node status and tumour thickness remain the dominant but imperfect guides to management. The Dalian team frames its letter as constructive engagement, crediting the original authors with a robust cohort and a rigorous multivariate analysis in the overall population, while urging that subgroup-level claims be contextualised with the statistical caution they require.

The broader lesson radiates well beyond melanoma genomics. Modern high-throughput studies routinely generate dozens or hundreds of candidate associations, and the path from an exploratory signal to a clinically actionable biomarker runs through validation in independent cohorts, replication under pre-specified analytical plans and harmonisation with existing staging and risk models. A deletion at 11q23.1-3 affecting CBL may yet prove to be a genuine driver event with prognostic power, and the original study’s evidence in the overall cohort suggests the hypothesis is worth pursuing vigorously. But as Han, Gao and Jiang make clear, the credibility of that pursuit depends on how the statistics are handled at each step, and on whether the field resists the temptation to treat a subgroup p-value of 0.044 as a verdict rather than a prompt for further, more stringently powered investigation.

For patients with stage II melanoma, the stakes are concrete: biomarkers of this kind could ultimately refine who is monitored most intensively, who is considered for adjuvant intervention and who can be reassured. Ensuring that such tools rest on statistically sound foundations is therefore not an academic quibble but a patient-safety issue. The correspondence, received on 5 June 2026, revised on 14 June and accepted on 3 September before publication on 21 September, stands as a reminder that in precision oncology, the rigour of the analysis is inseparable from the value of the discovery, and that the most important filters between a genomic observation and a clinical biomarker are multiple-testing correction, confounder control and disciplined adherence to reporting guidelines such as REMARK.

Subject of Research: Methodological evaluation of the 11q23.1-3 CBL deletion as a prognostic biomarker in stage II melanoma

Article Title: Methodological considerations in defining CBL as a prognostic biomarker in stage II melanoma

Article References: Han, J., Gao, X., & Jiang, W. (2026). Methodological considerations in defining CBL as a prognostic biomarker in stage II melanoma. British Journal of Cancer, 135(9), 1357-1358. https://doi.org/10.1038/s41416-026-03628-2

Image Credits: AI Generated

DOI: 10.1038/s41416-026-03628-2

Keywords: stage II melanoma, CBL, 11q23.1-3 deletion, prognostic biomarker, RAS-mutated melanoma, multiple testing adjustment, REMARK guidelines, subgroup analysis, relapse-free survival, AJCC staging, melanoma genomics, statistical methodology

Melflufen, also known as melphalan flufenamide, is a peptide-drug conjugate designed to deliver an alkylating payload preferentially into myeloma cells by exploiting peptidase activity within the tumor microenvironment. The OCEAN trial compared melflufen-dexamethasone against pomalidomide-dexamethasone in patients with relapsed or refractory disease, while HORIZON evaluated melflufen in heavily pretreated patients. The current European approval restricts melflufen use to patients who have not undergone autologous stem cell transplant or whose time to progression is three years or more, a restriction grounded in a post-hoc analysis of OCEAN showing that patients with a time to progression under 36 months fared significantly worse on melflufen than on the comparator, with median overall survival of 15.7 versus 28.7 months and a hazard ratio of 1.8 that reached statistical significance.

The commentators focus their first major criticism on Patient 1 from the case series, whose time to progression after autologous stem cell transplant was 28 months, placing this individual squarely within the subgroup that the OCEAN post-hoc analysis identified as being disadvantaged by melflufen treatment. The original paper acknowledged this discrepancy only briefly, noting that the long duration of response occurred despite the unfavorable time to progression. Bhise and colleagues contend that this framing creates a subtle but consequential problem: a single case with an atypically favorable outcome is being used to soften the message about a subgroup that randomized trial evidence indicates is harmed, on average, by this treatment relative to the alternative. They emphasize that case reports are inherently susceptible to selective emphasis because exceptional outliers are precisely what prompt such reports in the first place.

The statistical logic underlying this criticism reflects a well-known pitfall in clinical evidence interpretation. Subgroup analyses of randomized trials estimate average effects within defined patient populations, and individual outcomes within any subgroup can deviate dramatically from that average without invalidating the population-level finding. When a case report highlights an outlier who thrived despite belonging to a subgroup that trial data show is disadvantaged, presenting that case alongside arguments for extending treatment to patients outside the approved population risks leaving readers with the impression that individual favorable responses can offset subgroup-level harm demonstrated in a randomized comparison. The commentators argue that explicitly restating, within the discussion of this specific case, that the OCEAN subgroup data indicate a significant survival disadvantage for similar patients would prevent the exceptional case from being misread as evidence against the very population-level finding it contradicts.

The second major criticism concerns the safety data used to reassure readers about the risk of second primary malignancies. Patient 1 received 46 cycles of melflufen-dexamethasone over approximately four years before developing a second primary myelodysplastic neoplasm, a type of bone marrow disorder that can progress to acute leukemia. The original authors attributed this malignancy largely to the patient’s prior exposure to alkylating agents and immunomodulatory drugs, citing the low reported rate of second primary malignancies across the HORIZON and OCEAN trials as reassurance about melflufen’s own mutagenic contribution. The commentators find this reassurance statistically unsound for a specific and important reason rooted in the relationship between cumulative exposure and carcinogenic risk.

That reason is exposure duration. The median progression-free survival among trial responders was 8.5 months in both study populations, meaning that most trial patients received only a small fraction of the cumulative melflufen exposure that Patient 1 accumulated before the malignancy developed. Cumulative-dose-related carcinogenesis is a well-established feature of alkylating agents, and the commentators point out that the trial population’s exposure was systematically too brief to detect such a risk even if it existed. A second primary malignancy rate calculated from a population whose treatment was, on average, cut short after several months by disease progression or intolerance therefore provides limited reassurance about mutagenic risk in patients who go on to receive several years of continuous alkylating exposure. Because the original discussion favors considering melflufen for extended use in exceptional responders, the very population capable of accumulating multi-year exposure, the safety reassurance drawn from short-exposure trial populations does not address the risk profile most relevant to the patients the recommendation would actually affect.

The commentators propose a concrete methodological remedy: reporting cumulative melflufen dose or treatment duration alongside second primary malignancy incidence, where such data are available from trial extensions or real-world follow-up, would provide a more directly applicable estimate of risk for long-term responders being considered for extended therapy. This recommendation aligns with broader principles in oncology pharmacovigilance, where the relevance of safety signals depends critically on matching the exposure profile of the safety data to the exposure profile of the patients in question. Similar concerns about second primary malignancies have been raised in other novel therapy contexts, including systematic reviews of malignancy risk following CAR T-cell therapy, underscoring that long-duration follow-up is essential whenever powerful cytotoxic or immunologic treatments achieve durable disease control.

Despite these pointed criticisms, the commentary is not a rejection of the original work. Bhise and colleagues explicitly acknowledge that the case series offers a genuinely useful clinical contribution by illustrating that meaningful, durable responses to melflufen-dexamethasone are achievable in carefully selected patients. They note that such granular, longitudinal case detail complements population-level trial reporting in ways that aggregate statistics cannot, capturing the clinical course of individual patients over years of treatment. They also credit the original authors with appropriate transparency for explicitly acknowledging that the cases were retrospectively selected for their exceptional duration of response, an admission that properly frames the study’s inherent selection bias. This balanced tone reflects a growing recognition in the medical literature that case reports retain value precisely when their limitations are stated plainly and their conclusions are calibrated to the strength of the evidence.

The clinical implication drawn by the commentators is direct and specific. Before these exceptional cases inform practical guidance about extending melflufen use beyond the currently approved target population or into multi-year continuous therapy, two reconciliations are needed. First, the discussion of Patient 1 should explicitly confront the unfavorable OCEAN subgroup data for patients with a time to progression under 36 months, rather than allowing a favorable outlier to stand in implicit tension with randomized evidence. Second, reassurance about second primary malignancy risk during extended use should be grounded in exposure-matched safety data rather than incidence rates drawn from a trial population whose average treatment duration was substantially shorter. Until those conditions are met, the commentators suggest, the exceptional responses should be viewed as intriguing observations rather than as a basis for broadening treatment recommendations.

The exchange highlights a broader lesson for the era of precision oncology and expanding immunotherapies. As treatment landscapes grow more complex, with novel agents such as antibody-drug conjugates, bispecific antibodies, and CAR T-cell therapies entering practice alongside older cytotoxic backbones, the temptation to generalize from spectacular individual responses will only increase. The commentary by Bhise, Akotkar, Bhandari, and Gite serves as a reminder that the evidentiary weight of a case report is bounded by the statistics of the trials that contextualize it, and that safety conclusions are only as strong as the exposure profiles on which they rest. For patients with relapsed or refractory multiple myeloma, a disease where treatment sequencing decisions carry profound survival consequences, the difference between an exceptional anecdote and population-level evidence can be a matter of years of life, making rigorous interpretation of both not an academic nicety but a clinical imperative.

Subject of Research: Critical appraisal of melflufen-dexamethasone case reports for relapsed/refractory multiple myeloma

Article Title: Comment on “Exceptional long-term responses from OCEAN and HORIZON trials: melflufen-dexamethasone as an expansion of treatment options for relapsed/refractory multiple myeloma in the era of new immunotherapies?”

Article References: Bhise, M. R., Akotkar, A., Bhandari, S., & Gite, K. V. (2026). Comment on “Exceptional long-term responses from OCEAN and HORIZON trials: melflufen-dexamethasone as an expansion of treatment options for relapsed/refractory multiple myeloma in the era of new immunotherapies?”. Journal of Cancer Research and Clinical Oncology, 152(9), Article 176. https://doi.org/10.1007/s00432-026-06608-4

Image Credits: AI Generated

DOI: 10.1007/s00432-026-06608-4

Keywords: melflufen, dexamethasone, multiple myeloma, OCEAN trial, HORIZON trial, relapsed/refractory myeloma, second primary malignancy, subgroup analysis, case report, alkylating agents, clinical trials, oncology