Can Digital Tools Effectively Monitor Neurotoxicity in CAR T-Cell Therapy?

Can Digital Tools Effectively Monitor Neurotoxicity in CAR T-Cell Therapy?

The clinical landscape of cellular immunotherapy is shifting toward more personalized and responsive monitoring strategies, yet the shadow of neurotoxicity remains a persistent hurdle for clinicians and patients alike. Efforts to create an aggregate cognitive score are frequently thwarted by the high treatment burden and constitutional symptoms, such as fever, that prevent patients from completing complex tasks. Chimeric antigen receptor (CAR) T-cell therapy has emerged as a transformative treatment for blood cancers, yet it carries the risk of immune effector cell-associated neurotoxicity syndrome (ICANS). This condition can range from minor speech issues to severe cerebral edema, making frequent monitoring essential. A recent pilot study at University Hospital Würzburg investigated whether a tablet-based application called “NeuroCAT” could offer more detailed bedside insights than the traditional, intermittent clinical checks currently in use. Traditional neurotoxicity assessments rely heavily on the Immune Effector Cell Encephalopathy (ICE) score, which is typically administered by nursing staff at fixed intervals. While effective for detecting overt changes, these manual checks may miss subtle or fluctuating symptoms that occur between scheduled evaluations. Digital tools are hypothesized to fill this gap by providing a continuous stream of data regarding a patient’s cognitive state and physical symptoms. By allowing patients to engage with tasks directly from their hospital beds, clinicians hope to identify the early warning signs of neurotoxicity with greater precision than manual scoring allows.

Clinical Study Framework

Participant Profile and Clinical Demographics

The pilot study enrolled 33 adult candidates between January and August, focusing specifically on those diagnosed with relapsed or refractory hematologic malignancies. The group was primarily composed of individuals battling multiple myeloma, which accounted for over 72 percent of the cohort, while the remaining participants were treated for diffuse large B-cell lymphoma. With a mean age of approximately 60.7 years, these participants represented a demographic that researchers initially feared might struggle with sophisticated digital interfaces during periods of intense medical distress. However, the initial response was overwhelmingly positive, yielding a 100 percent participation rate among those eligible and approached for the study. This high level of initial engagement suggests that patients facing life-threatening illnesses are highly motivated to contribute to research that might improve the safety of their treatments. The predominantly male cohort provided a stable baseline for evaluating how digital tools could be integrated into the rigorous environment of an inpatient oncology ward, where distractions and physical discomfort are constant throughout the recovery process.

Implementation of Methodology and Phases

The research protocol was meticulously divided into three specific phases to capture the full spectrum of the patient experience: a baseline assessment (T0) before treatment, an acute phase (T1) during the peak of hospitalization, and a final evaluation (T2) prior to discharge. This structure allowed researchers to track changes across the most critical windows of the therapeutic process, particularly during the high-risk period for cytokine release syndrome and neurotoxicity. The baseline assessment was performed prior to lymphodepletion, ensuring that the initial data represented the patient’s cognitive status without the confounding influence of chemotherapy or the modified T-cell infusion. By establishing this clear starting point, the study aimed to measure the specific impact of the cellular therapy itself on cognitive function and motor skills. This longitudinal approach is essential for understanding how symptoms evolve in real time, as traditional one-time tests often fail to capture the transient nature of neurotoxic events that can occur and resolve between physician rounds.

NeuroCAT Platform Design and Parameters

NeuroCAT was developed as a comprehensive digital suite featuring twenty-five cognitive and visual tasks, alongside a survey covering nine symptom domains. The tasks were designed to mirror classic neuropsychological tests that measure attention, memory, and motor coordination, including digital adaptations of the Trail Making Test and Stroop interference tasks. By migrating these assessments from paper-and-pencil formats to a tablet-based interface, the platform provides a more dynamic way to capture performance metrics. Each task was designed to be intuitive, allowing patients with varying levels of digital literacy to navigate the interface with minimal assistance from hospital staff. The visual nature of the tasks aims to bypass language barriers and education-related biases that often skew traditional verbal tests. This design philosophy emphasizes the need for bedside tools that are accessible and robust enough to function in a loud, busy hospital environment while maintaining the scientific integrity of the underlying neuropsychological principles they are intended to measure.

Tailored Task Selection for Stamina

To account for the physical exhaustion and fevers often associated with cellular therapy, clinicians used a “tailored” approach during the follow-up phases of the study. Rather than requiring the full 30-minute battery of 25 tasks, they selected a subset of roughly eight tasks per session to match each patient’s immediate stamina and baseline performance. This flexibility was crucial because patients undergoing CAR T-cell therapy frequently experience profound fatigue and systemic inflammation that make prolonged cognitive effort impossible. By reducing the task load, investigators ensured that participants could remain engaged without becoming overwhelmed or discouraged by the testing process. However, this individualized tailoring introduced a significant methodological trade-off, as it prioritized participant adherence over the consistency of the data collected. While this allowed the study to maintain high participation rates, it meant that not every patient performed the same tasks at every time point, complicating the process of comparing results across the entire cohort. This highlights the ongoing tension between creating patient-friendly clinical tools and maintaining the rigorous standardization required for diagnostic validity.

Analysis of Feasibility and Data Accuracy

Completion Rates and Longitudinal Adherence

The study reported a remarkably high completion rate, with over 90 percent of participants finishing all three assessment stages despite the rigors of their treatment. This success suggests that digital tools are indeed feasible in an acute hospital setting, provided the interface is user-friendly and the time commitment is manageable for those in recovery. Of the 33 patients who began the study, 30 completed the final assessment, with the remaining three dropping out due to death or severe clinical instability unrelated to the app’s functionality. This level of adherence is significant because it proves that elderly cancer patients are willing to use digital health platforms even when they are feeling profoundly unwell. The investigators noted that the presence of a supportive clinical team likely contributed to these high rates, as patients felt that their participation was a meaningful part of their overall care plan. This finding underscores the potential for digital monitoring to become a routine part of the oncology workflow, as long as it is introduced with clear communication and integrated seamlessly into the existing hospital infrastructure.

Evaluation of Feasibility in Distress

While the high adherence rates suggest that digital tools are feasible, the researchers noted that this success was largely due to the flexibility of the shortened follow-up sessions. The findings do not necessarily prove that a comprehensive cognitive battery is sustainable for patients in significant distress; instead, they highlight that for a digital tool to remain viable, it must be adaptable and brief enough to accommodate fluctuating clinical status. During the acute phase, many patients likely would have failed a full assessment if the investigators had not intervened to prune the task list. This reveals a fundamental truth about bedside monitoring: the “feasibility” of a tool is highly dependent on its ability to be modified on the fly to meet the patient’s current physiological capacity. If a digital platform is too rigid, it risks high dropout rates precisely when the most critical data needs to be collected—during the onset of neurotoxicity. Future versions of such tools will likely need automated algorithms that can adjust task difficulty and duration based on real-time performance and physiological markers like heart rate or temperature.

Challenges in Statistical Interpretation and Data

Despite high engagement, the quantitative data revealed significant methodological hurdles, such as “ceiling effects,” where tasks were too easy, and “floor effects,” where they were too difficult for the target population. For instance, tasks involving simple object naming or basic orientation resulted in nearly all patients achieving perfect scores, providing very little insight into subtle cognitive fluctuations. Conversely, more complex divided-attention tasks were so difficult for some patients at baseline that they could not be used to track decline during the acute phase. These statistical extremes make it difficult to identify the “sweet spot” where a task is challenging enough to detect impairment but simple enough to be completed by someone with early-stage neurotoxicity. The lack of standardized scoring across various tasks further complicated the analysis, as there was no clear way to weigh the importance of a tremor in a drawing task against a delay in a memory task. This lack of uniformity is a primary barrier to transforming digital performance data into a clear clinical signal that can be used to guide treatment decisions.

Informative Missingness and Aggregate Scoring

Because the tasks in the follow-up sessions were nonrandomly selected by clinicians to avoid discouraging patients, the resulting data suffered from what researchers call “informative missingness.” This bias occurred because tasks that were already difficult for a patient at baseline were often omitted in subsequent sessions to prevent frustration. Consequently, the dataset was not missing at random, but rather because of the very cognitive deficits the study aimed to measure. This statistical bias made it impossible for researchers to create a standardized aggregate cognitive score that could be applied across the entire patient population. Without a uniform dataset across all participants, tracking a clear cognitive trajectory or establishing a reliable “digital biomarker” for neurotoxicity remains a significant challenge. To overcome this, future studies will need to implement a fixed core set of tasks that are mandatory for all participants, alongside a secondary set of optional, more difficult tasks to probe for higher-level cognitive changes in those who are feeling well enough to attempt them.

Patient Experience and Diagnostic Sensitivity

Monitoring Subjective Symptom Volatility and Trends

In addition to cognitive tasks, the digital platform tracked patient-reported symptoms such as fatigue, dizziness, and difficulty concentrating, revealing a high degree of volatility in individual trajectories. While some patients reported improvements during the acute phase—perhaps due to the intensive medical support they were receiving—others saw sharp declines that mirrored their clinical status. This symptom tracking provided a more granular view of the patient experience than a standard nursing assessment, capturing the “ebb and flow” of daily life in a hospital room. However, the data showed that subjective reports do not always correlate with objective cognitive performance. A patient might report feeling sharp and focused while simultaneously failing a memory task, or they might feel profoundly fatigued while maintaining high scores on attention tests. This discrepancy suggests that while patient-reported outcomes are valuable for understanding quality of life, they cannot be used as a standalone diagnostic tool for neurotoxicity. The volatility of these symptoms highlights the need for a multimodal approach that combines subjective reports with objective performance metrics.

Technical Hurdles in Patient-Reported Data

Technical issues also surfaced during the study, reminding researchers that early-stage digital prototypes often require significant refinement before they can be trusted for clinical data collection. For example, a mood screening item had to be entirely excluded from the final analysis because inconsistent data coding in the software’s output rendered the results uninterpretable. These types of logging errors are common in investigational apps but can be devastating for a small pilot study where every data point is crucial. Furthermore, the absence of precise timestamps or response latency data meant that “processing speed”—a key indicator of neurological health—was not captured with enough accuracy to be useful. Without the ability to measure how long a patient paused before making a decision, the researchers lost a critical dimension of cognitive assessment. This technical limitation emphasizes that digital health developers must prioritize back-end data integrity and high-resolution timing as much as they prioritize the user interface, as these granular details are where the early signs of neurotoxicity are most likely to be found.

Distinguishing ICANS from General Treatment Fatigue

A critical finding of the study was the tool’s limited ability to specifically identify ICANS among the broader symptoms of recovery from treatment. While patients diagnosed with neurotoxicity showed spikes in certain symptoms, many patients without ICANS reported similar increases in fatigue and sleep disruption. This overlap makes it extremely difficult to distinguish between the specific neurological effects of CAR-T cells and the general exhaustion that follows high-dose chemotherapy and immune activation. The cytokine storm that often precedes ICANS causes high fevers and systemic malaise, which can mimic the early signs of neurotoxicity on almost any digital task. Consequently, a drop in a cognitive score might reflect a patient’s high fever rather than a direct impact on their central nervous system. This lack of specificity is a major hurdle for any monitoring tool, as it risks triggering “false alarms” that could lead to unnecessary interventions or medical anxiety. Refining these tools will require identifying specific cognitive signatures that are unique to ICANS and distinct from general systemic illness.

Diagnostic Sensitivity of Graphomotor Assessment

Digital graphomotor tasks, such as drawing spirals on a tablet, exhibited qualitative irregularities in affected patients but lacked the standardization necessary for objective diagnosis. While researchers observed visible “shakiness” or line deviations in the drawings of patients who developed ICANS, these observations were not reviewed by blinded experts or compared against a standardized scale. Furthermore, the conditions under which the drawings were made—such as the patient’s position in bed or the use of a stylus versus a finger—were not strictly controlled, introducing significant noise into the data. Without a standardized method to quantify these motor deviations, the “spiral test” remained an interesting qualitative observation rather than a reliable clinical metric. To be useful in a diagnostic setting, these digital motor tasks must produce a numerical score that can be compared against a known baseline of healthy motor function. This would require advanced image-processing algorithms capable of calculating fine-motor tremors and pressure variations that are invisible to the naked eye but indicative of underlying neurological distress.

Future Requirements for Clinical Integration

Validating Tools for Cellular Immunotherapy

The investigators identified several foundational barriers that prevented digital tools from being integrated into standard clinical decision-making during the observation period. The small sample size and the lack of a control group limited the generalizability of the findings, while the absence of validation against “gold-standard” neuropsychological tests left the accuracy of the platform in question. Because the digital assessments were conducted on a fixed schedule rather than being triggered by clinical changes, it was difficult to determine if the app could actually detect neurotoxicity earlier than a human clinician would. The research team noted that for these tools to move from a research context into routine practice, they had to undergo a rigorous, staged validation framework. This process included verifying that the app’s algorithms were consistent with granular, raw data records and ensuring that every performance drop was clinically significant. Until such validation was achieved, the platform remained a supportive tool for documenting the patient experience rather than a primary diagnostic instrument for managing high-risk cellular therapies.

Advancing Digital Monitoring in Oncology

Ultimately, the pilot study at University Hospital Würzburg demonstrated that while patients were willing and able to engage with digital monitoring during CAR T-cell therapy, the technology was not yet a replacement for traditional clinical grading. The move toward “decentralized” bedside monitoring was found to be highly promising for capturing the subjective patient experience, but the study proved that feasibility was not a substitute for clinical utility. Future developments must focus on capturing more granular data points, such as response latencies and precise processing speeds, which are key indicators of cognitive health. Larger multicenter trials were recommended to establish “reliable change thresholds” that could alert medical staff to significant deviations in a patient’s status in real time. By bridging the validation gap through rigorous psychometric testing, the medical community moved closer to creating high-fidelity data points that truly supplemented expert judgment. The work concluded that while digital tools added a layer of granularity to patient monitoring, they required further technical and clinical refinement to become actionable biomarkers in the management of neurotoxicity.

Subscribe to our weekly news digest.

Join now and become a part of our fast-growing community.

Invalid Email Address
Thanks for Subscribing!
We'll be sending you our best soon!
Something went wrong, please try again later