Time in biotech trials: why six months do not equal two years | $CADL, $IOVA, $KOD, $KYTX

01A clinical result has several clocks

Six months of encouraging clinical data can establish that a treatment is doing something useful. They cannot automatically establish how that benefit behaves after two years. The missing interval contains opportunities for sustained improvement, relapse, delayed toxicity, treatment changes and incomplete observation. Time therefore changes the question a study can answer, even when its headline response rate barely moves.

There is a patient's clock, beginning at a precisely defined event such as randomization, infusion or first response. There is a study clock, covering recruitment, visits and the final date included in an analysis. There is a regulatory clock, during which a sponsor assembles evidence and an agency evaluates a particular application. Finally, there is a company's financial clock: salaries, manufacturing commitments and clinical obligations continue while observations accumulate. These clocks overlap without becoming interchangeable.

The distinction is particularly visible in Candel Therapeutics, Iovance Biotherapeutics, Kodiak Sciences and Kyverna Therapeutics. A prostate cancer recurrence, the end of a melanoma response, a change in visual acuity and a walking assessment in stiff person syndrome are different events. A month contributes differently to each analysis. Putting the four companies on a common calendar is useful; putting their outcome percentages on a common performance ranking would be misleading.

The central task is to identify what additional observation can resolve. More time may confirm persistence, expose late failures or produce enough events to estimate an effect with greater precision. It cannot retrospectively create randomization, recover every missing measurement or turn an exploratory subgroup into a prespecified primary endpoint. A study can become older without becoming equally informative on every question.

For an investor, this makes data maturity a practical tool. It separates evidence already observed from the period still being forecast. For a reader interested in the medicine, it explains why a promising treatment can deserve further development while its long-term benefit remains uncertain. Neither judgment requires dismissing early results. Both require respecting the boundary of observation.

02Four companies, four different meanings of time

Candel's localized prostate cancer trial asks whether adding an investigational treatment to radiotherapy changes the occurrence of defined disease events. These events can take years to accumulate. A long period without recurrence matters, but the exact definition of recurrence and the number of events remain central. Biopsy findings, biochemical failure, metastasis and death are related outcomes, not synonyms.

Iovance provides a different setting. Lifileucel is an autologous tumor-infiltrating lymphocyte therapy marketed as Amtagvi for a defined previously treated melanoma population in the United States. Its clinical evidence includes patients whose tumors responded and whose responses were subsequently followed. The duration among responders and survival among everyone analyzed have different denominators. A very long response in one patient is meaningful without describing the typical experience of every treated patient.

Kodiak's DAYBREAK study evaluates visual outcomes under specified dosing regimens. Its primary analysis concerns change in visual acuity averaged across late visits, rather than waiting for half the participants to experience a cancer event. The interval between injections is another measure again. A person can maintain vision with additional injections; that outcome may support efficacy while changing the interpretation of treatment burden.

Kyverna's KYSA-8 study follows functional change after investigational miv-cel in stiff person syndrome. Its September 24 update extends observation beyond the original 16-week efficacy assessment. Sustained walking improvement and time without chronic immunotherapy are relevant but distinct outcomes. Neither can be translated directly into a cancer response rate or a retinal dosing interval.

These differences determine what a later update must contain. Candel needs endpoint-specific event accounting. Iovance needs response duration together with the original response denominator. Kodiak needs the scheduled vision analyses and the actual treatment received. Kyverna needs paired functional observations, concomitant treatments and continued safety follow-up. The shared principle is precise: the clock must remain attached to a population, an outcome and a treatment strategy. Without those three anchors, a duration is an attractive number with an unclear meaning.

03Publication date, data cut and database lock

A press release dated September does not necessarily contain observations through September. The data cut defines the latest clinical information eligible for an analysis. Database lock comes after the processes used to clean, reconcile and finalize the relevant records. A conference presentation may appear later still. The calendar on the announcement and the calendar inside the dataset answer different questions.

Kyverna makes this distinction unusually clear. Its September 24, 2026 release describes a July 2026 database lock for the one-year stiff person syndrome analysis. Candel's May 2026 extended prostate cancer update uses a March 15, 2026 cut. The interval between the clinical cut and public presentation should not be added to the stated follow-up. A reader cannot give a participant two extra months of documented observation merely because the announcement arrived later.

There is also a difference between additional follow-up in the same participants and an enlarged population. An update may add newly enrolled patients, extend existing patients' observation, or do both. If the first cohort has two years of follow-up and many later patients have only a few months, the combined response rate may look stable while the evidence for two-year durability still rests on a much smaller group.

The comparison between successive releases therefore starts with membership. Are the same cohorts included? Has the treatment dose changed? Are responses confirmed under the same rules? Is the analysis based on all treated patients, an efficacy-evaluable subset or only responders? The answer can explain an apparent change before any biological interpretation is necessary.

A revised dataset is not inherently suspicious. Clinical records mature, central reviews conclude and previously pending scans become available. Transparency means being able to distinguish those ordinary processes from a genuine change in the treatment's observed performance. The most informative update supplies both its cut date and the path from the previous dataset to the new one. Kyverna's SEC-furnished release provides the current example.

04Median follow-up is not a minimum guarantee

The phrase “median follow-up of two years” is often heard as though every participant had been observed for two years. That is incorrect. A median summarizes a distribution, and the method used to calculate the distribution matters. Some participants may have substantially shorter observation, while a smaller early-enrolled group contributes much longer experience.

The methodological literature has recognized this problem for decades. Schemper and Smith's 1996 paper explains that different methods of quantifying follow-up can produce materially different summaries. A simple median of observed times, a median among surviving participants and a reverse Kaplan–Meier estimate do not necessarily describe the same thing. The method should be named before two reports are treated as directly comparable. Schemper and Smith, Controlled Clinical Trials.

A hypothetical example illustrates the limitation. A study recruits over eighteen months and publishes shortly after the final participant reaches month six. Its earliest participants may have two years of observation. The study genuinely contains two-year data, but it does not contain two-year data for the entire cohort. A statement about the longest observed duration cannot replace a statement about how many participants reached the clinically important time point.

The most useful companion measures are the minimum observation where available, the distribution of follow-up, the number still being assessed at key visits and the number contributing to each endpoint. These answer practical questions that one median cannot. They show whether a later interval reflects broad experience or a handful of early participants.

Longer follow-up can also change the composition of the people still observed. Some have already experienced an event; others have withdrawn, died from another cause or simply have not yet reached the later visit. These reasons carry different implications. A sophisticated follow-up calculation improves description of the dataset, but it does not eliminate the need to understand why observation ended for individual patients.

05Censoring preserves partial information

In a time-to-event analysis, a censored observation usually means that the event of interest had not been observed by the last usable time for that participant. The person still contributes information up to that point. Censoring does not mean the person failed treatment, and it does not mean the person is guaranteed to remain event-free afterward.

Consider a purely illustrative participant whose cancer has not progressed at a month-nine assessment when the analysis closes. The study knows something useful: no qualifying progression was observed during that interval. It does not know the participant's month-eighteen outcome from that record. Counting that person as an eighteen-month success would invent information. Discarding the entire nine months would throw information away. Survival methods exist partly to handle this middle ground.

Administrative censoring at a common data cut is different from losing a patient because worsening health prevents further visits. Standard analyses rely on assumptions about the relationship between censoring and the event process. If people at higher risk disproportionately disappear, the remaining observations may give an overly favorable picture. The issue is the reason for missing observation, not the mere presence of censoring marks on a graph.

Treatment changes create another layer. A new anticancer therapy, rescue injection or return to immunosuppressive treatment can alter what subsequent outcomes mean. The protocol must specify how such events relate to the question being estimated. The ICH E9(R1) estimand framework, adopted in FDA guidance, emphasizes this connection between the clinical question, events after treatment starts and the analysis strategy.

For the four companies, the practical consequence is straightforward. A later update is stronger when it explains losses to follow-up and changes in treatment, rather than reporting only an attractive percentage among those still measurable. Good retention and transparent accounting make extended observation more credible. A long calendar interval with unclear patient disposition can leave the central uncertainty largely intact.

06The number at risk tells the story behind the curve

A Kaplan–Meier curve may extend well beyond the point where most participants remain under observation. The line continues because the method can estimate survival using the remaining information. Its visual length should not be mistaken for equally strong evidence at every point along the horizontal axis.

The accompanying number-at-risk table shows how many participants remain event-free and under observation just before specified times. Later numbers usually shrink as events occur, observations end or participants have not yet accumulated sufficient time. A late plateau based on few people has a different evidentiary weight from a plateau supported by a large, well-followed group.

This is not a cosmetic concern. In a simplified example, one additional event among a large risk set changes the curve less than one event among a very small risk set. Confidence intervals generally reveal some of that loss of precision. The late curve can therefore move substantially when only a few new observations arrive, even if the overall study contains hundreds of participants.

Gebski, Garès, Gibbs and Byth examined this distinction in their methodological paper on data maturity and follow-up. Their work addresses how the remaining observations and the effect of additional events inform the maturity of time-to-event results. It supports treating maturity as more than elapsed time. Data maturity and follow-up in time-to-event analyses, International Journal of Epidemiology, 2018; DOI 10.1093/ije/dyy013.

For a reader comparing successive presentations, the useful question is whether the tail has gained support. Have more participants reached that interval? Has the number of events increased? Have confidence intervals narrowed? Has the apparent plateau survived broader observation? These questions can reveal substantial progress even without a more dramatic headline. Conversely, another presentation of the same thin tail may add little knowledge, despite a later conference date and a longer maximum follow-up.

07“Median not reached” is a boundary, not infinity

For a conventional Kaplan–Meier duration analysis, the median is the time at which the estimated proportion remaining without the event reaches one half. If the curve has not reached that threshold during available follow-up, the median may be reported as not reached. That result can be encouraging, but it does not specify the eventual median and certainly does not establish permanent benefit.

Two hypothetical studies can both report an unreached median for very different reasons. One may have followed many responders for several years with few relapses. Another may have enrolled recently, leaving most responders with short observation. The words are identical; the amount of supporting evidence is not. The number at risk, confidence interval, minimum follow-up and duration distribution separate these situations.

Iovance's November 3, 2025 NSCLC update provides a concrete example of why the accompanying details matter. The sponsor reported ten responses among 39 patients, including one response still awaiting confirmation, and an unreached median response duration with median follow-up of 25.4 months. These are dated interim findings in a particular lung cancer cohort. They are not the same dataset as the larger melanoma publication or an approved NSCLC indication. Iovance's original NSCLC announcement.

An unreached median can later become a finite value as events accumulate. That change is not automatically evidence of deterioration relative to expectations. It can simply mean the analysis has become more informative. The relevant comparison concerns the full curve, observed durations and clinical context, not whether a once-open field in a table now contains a number.

Similarly, a median that remains unreached is not automatically improving. If little new observation has accumulated, uncertainty may remain broad. The informative development is the addition of well-characterized time and patients, rather than the continued presence of the same two letters, NR, in successive slides.

08Response duration belongs to responders

Objective response rate asks how many people met defined tumor-response criteria in the specified analysis population. Duration of response asks how long those responses lasted, generally beginning when response was first documented. The second calculation conditions on having responded. It should never quietly replace the first denominator.

Suppose, only for illustration, that thirty of one hundred treated patients respond and the response-duration analysis is favorable. That does not mean all one hundred experienced the reported duration of benefit. The remaining patients may have stable disease, early progression, missing assessments or another outcome under the study rules. The full clinical picture requires the response rate, the duration among responders and outcomes covering the wider population.

Time to first response matters too. A response that appears later contributes a shorter measured response duration at the same data cut, even if the patient has been on study for a long time. This explains why follow-up from infusion and follow-up from response cannot always be substituted. The starting point changes the amount of time relevant to the endpoint.

Confirmation rules add another distinction. An initial tumor reduction that qualifies at one assessment may require a subsequent assessment before it is counted as confirmed. A sponsor may present both confirmed and unconfirmed results, but the labels must remain visible. A later increase in confirmed response rate can reflect confirmation of existing observations rather than a completely new set of responses.

The FDA's oncology endpoint guidance discusses response magnitude and duration in the context of study design and disease setting. The investment implication follows directly: a durable responder population can create substantial clinical and commercial value, while the proportion of patients reaching that state remains crucial. A treatment with deep benefit for a subset and a treatment with modest benefit across most patients present different adoption, access and evidence questions. Neither can be understood through duration alone.

09Iovance: five-year observation reveals more than a response headline

The peer-reviewed five-year analysis of C-144-01 offers a useful example of what time can add. At the November 20, 2024 cut, the publication reported an objective response rate of 31.4%, a median response duration of 36.5 months and five-year overall survival of 19.7%. Among responders, 31.3% completed the five-year assessment with an ongoing response. The paper was published in the Journal of Clinical Oncology in 2025. Medina and colleagues; DOI 10.1200/JCO-25-00765.

The distinct denominators are the lesson. The response-duration statistic describes responders; the overall-survival statistic describes the analyzed population. The proportion of responders with an ongoing response at the five-year assessment is not a five-year response rate among everyone treated. Keeping these statements separate preserves the achievement without inflating its reach.

The analysis also reported responses that deepened over time. A later transition from partial to complete response can enrich understanding of treatment behavior. It does not erase early failures elsewhere in the cohort. Long follow-up makes both patterns visible: durable benefit in some patients and a different course in others.

Regulatory evidence uses its own defined population. FDA's original 2024 accelerated-approval summary described 73 patients treated within the recommended dose range in the primary efficacy cohort, with an ORR of 31.5%. That number should not be casually treated as a rounding variant of every later pooled publication. Cohort definitions and analysis sets must travel with the result. FDA's lifileucel approval summary.

The broader economic interpretation is conditional. Mature durability evidence can inform physician confidence and discussions of treatment value, but it does not by itself determine referral volume, manufacturing completion or reimbursement. Iovance's commercial results and clinical follow-up describe connected parts of the business. One measures recognized sales during a quarter; the other measures patient outcomes across years. A stronger understanding comes from relating them without pretending they are the same evidence.

10Candel: a long follow-up can leave a late endpoint immature

Candel's extended phase 3 prostate cancer analysis is especially instructive because the calendar is already long. The May 15, 2026 update reported a median follow-up of 58 months in a randomized trial of 745 patients. For prostate cancer-specific disease-free survival, the reported hazard ratio was 0.61, with a 95% confidence interval of 0.44–0.85. That result concerns a defined composite endpoint, not an established reduction in every later clinical outcome. Candel's AUA update.

The same release reported eight metastases among 496 participants in the investigational arm and seven among 249 in the control arm. The time-to-metastasis hazard ratio was 0.58, with a wide confidence interval of 0.21–1.59. These results illustrate how several years of observation can coexist with limited precision for a relatively infrequent event. The point estimate favors treatment, while the interval leaves substantial uncertainty.

The appropriate interpretation does not require choosing between enthusiasm and skepticism. The disease-free survival evidence can be favorable while the metastasis analysis remains immature. Different endpoints accumulate information at different speeds within the same randomized study. A future metastasis update may be valuable because it adds events and observation, not merely because its date is later.

The June 2026 publication announcement reports the original primary DFS result separately: HR 0.70, 95% CI 0.52–0.94. It also identifies prostate cancer-specific DFS and a post-hoc biopsy analysis. Comparing the primary DFS figure directly with the later prostate cancer-specific DFS figure as if treatment efficacy “rose from 30% to 39%” would mix endpoint definitions and analysis times. Candel's publication announcement.

For CADL, the useful progression is therefore a sequence of specific questions: persistence of the established signal, maturity of additional outcomes, regulatory review and potential clinical adoption. The sequence cannot be compressed into one percentage that grows with every presentation.

11Hazard ratios and absolute outcomes answer different questions

A hazard ratio compares event rates over time under the statistical model used. It is not a percentage-point difference in the proportion of patients who experience an event by a particular year. It is also not a direct statement that each treated patient gains the same amount of time. The NCI definition of hazard ratio provides a useful starting point, but interpretation still depends on the endpoint and analysis assumptions.

In a hypothetical example, a relative effect could accompany a change in an event probability from ten percent to seven percent, or from forty percent to twenty-eight percent. The relative changes look similar, while the absolute differences are three and twelve percentage points. These examples are arithmetic illustrations, not estimates for any of the four companies. They show why clinical significance needs baseline risk and a specified time horizon.

An absolute survival or event-free estimate at a given time also needs adequate support. If few participants have reached that time, a precise-looking percentage can sit inside a broad confidence interval. Presenting the estimate without its uncertainty can make a weakly supported tail appear much firmer than it is.

The proportional-hazards assumption deserves attention when a single HR summarizes a long interval. Treatment effects may emerge gradually or differ between early and late periods. A summary statistic can remain useful, but the curves and other planned analyses help explain what it averages. A visual crossing or delayed separation is a prompt to examine the model, not permission to select whichever segment looks most favorable.

For Candel, this means preserving the distinction between the reported time-to-event analyses and simple event counts. For Iovance, it means avoiding comparisons of single-arm median outcomes with an unrelated trial as though they produced a randomized hazard ratio. For every company, the most faithful description combines relative effect, absolute outcome, uncertainty and observation period whenever the available evidence supports all four.

12Primary, secondary and exploratory results age differently

An endpoint hierarchy establishes which questions a study is designed to answer formally and how the risk of false-positive conclusions is controlled. Additional months of follow-up do not automatically move an exploratory endpoint to the top of that hierarchy. A striking later subgroup result may remain exploratory even if its nominal p-value is small.

This is particularly important when a trial measures several clinical outcomes, analyzes several populations and reports them repeatedly over time. Each view can be informative, but the collection creates many opportunities to find an attractive pattern. FDA's guidance on multiple endpoints explains why multiplicity must be addressed rather than ignored.

A prespecified analysis with controlled error and an exploratory analysis developed after seeing the data serve different purposes. The first can test a defined hypothesis under the study's statistical plan. The second may generate a credible hypothesis, illuminate biology or guide a future trial. It becomes misleading only when the evidentiary distinction disappears from the description.

Candel's extended follow-up demonstrates the practical need for careful labels. The trial contains primary, secondary and exploratory outcomes, and its intermediate-risk subgroup is not interchangeable with the entire randomized population. More favorable numbers in a subgroup cannot simply replace the intention-to-treat result. They may provide useful context while needing independent confirmation or additional maturity.

Repeated looks at the same study also require care. A planned interim analysis is not equivalent to repeatedly checking an endpoint until it crosses a familiar significance threshold. The statistical design determines how interim information is handled. For investors, the consequence is that “another positive update” can represent several different developments: confirmation of a prespecified result, exploratory support, a newly mature endpoint or a repackaging of existing observations. The label matters because each development changes uncertainty in a different way.

13Kyverna: one-year persistence is a meaningful extension

Kyverna's September 24, 2026 release reports twelve-month data in 26 adults with stiff person syndrome in KYSA-8, a single-arm phase 2 registrational study of miv-cel. Median improvement in the timed 25-foot walk was reported as 46% from baseline at week sixteen and 49% at month twelve. The update extends the period over which functional benefit has been observed. It is not a randomized comparison with another treatment. Kyverna's clinical release filed with the SEC.

The frequently repeated 95% figure requires its full denominator: it refers to participants who had achieved the defined clinically meaningful walking improvement at the primary analysis and sustained that benefit. It does not mean that 95% of every possible patient with the disease responded. Nor does it mean the entire cohort achieved every secondary endpoint.

The extension from sixteen weeks to one year matters because an early functional improvement might otherwise prove temporary. Continued benefit broadens the evidence about persistence. At the same time, one year does not establish lifetime remission, and a small cohort cannot precisely characterize rare risks. Those limits coexist with an encouraging observation.

Time without chronic immunotherapy supplies another perspective on treatment burden, but it should remain attached to its definition. Resuming a treatment, reducing its dose and remaining completely off it describe different experiences. Future reports become more useful when they describe the entire trajectory, including rescue interventions and the reasons for them.

Kyverna planned to include the one-year data in a rolling BLA targeted for completion in the fourth quarter of 2026. Submission, acceptance for review and approval are separate future milestones. For the financial interpretation, the one-year update can strengthen the evidence package without eliminating manufacturing, regulatory and commercial questions. Its value is the replacement of a previously unobserved interval with actual follow-up. The next interval still requires observation.

14Kodiak: a scheduled visit is different from an event-driven clock

Kodiak's DAYBREAK design highlights a different kind of maturity. The study compares investigational Zenkuda and KSI-501 regimens with aflibercept in treatment-naïve wet age-related macular degeneration. The stated primary endpoint is non-inferiority in change in visual acuity from baseline to the average of weeks forty, forty-four and forty-eight. Kodiak's August update had forecast September topline results. On September 25, the company scheduled the release for September 28, 2026, with a webcast at 8:30 AM ET. As of this article’s September 26 cut-off, that announcement establishes the presentation date, not the trial outcome. Kodiak’s dated announcement. Kodiak's second-quarter update.

Here, the main clock concerns participants reaching scheduled assessments and the completeness of those measurements. A study does not become ready merely because an early participant reaches the final visit. Later enrollees, missing visits and data reconciliation affect when a sufficiently complete analysis can be produced. This differs from an event-driven cancer analysis that waits for a specified number of events.

Non-inferiority also has a specific meaning. It tests whether the result excludes an unacceptable loss of efficacy relative to an active comparator under the chosen margin and analysis plan. It does not establish that two therapies are identical, and it is not automatically evidence of superiority. The margin, confidence interval and sensitivity analyses matter. FDA's non-inferiority guidance.

Durability introduces a second question alongside vision. How much treatment was needed to achieve the measured visual outcome? If additional dosing is allowed, the observed efficacy belongs to the entire assigned strategy, including those permitted interventions. A maximum permitted interval between injections is a design feature; it is not proof that every patient can maintain that interval.

For KOD, the informative readout therefore needs both axes: preserved visual function under the study's analysis and the distribution of actual treatment burden. A favorable late vision average with frequent intervention and a favorable average with less intervention could have different practical implications. The time printed in a trial name or headline cannot resolve that difference.

15Durability must describe a patient experience

“Durable” is meaningful only when the benefit being sustained is clear. A sustained tumor response, a period without recurrence, preserved vision and improved mobility describe different experiences. Even within one disease, laboratory change, symptom relief and freedom from additional treatment may have different trajectories.

For a functional measure, repeated assessments help distinguish persistent benefit from a particularly favorable visit. Measurement conditions, assessor training and the handling of missed visits influence interpretation. A numerical improvement may be statistically persuasive while its practical significance depends on the baseline level and what the patient can now do. Walking more easily, maintaining independence and reducing treatment burden are related but separate observations.

For a chronic ocular disease, longer intervals can reduce appointment and injection burden. However, the full schedule matters: loading doses, monitoring visits, rescue treatment and subsequent interval adjustments all contribute to the patient's experience. A six-month maximum dosing interval does not necessarily mean only two healthcare encounters in a year. The economic implications for providers and patients depend on the complete care pathway.

For a one-time cellular treatment, durability is measured after a concentrated treatment episode that can include preparative and supportive care. The absence of routine retreatment does not mean the treatment pathway had no burden. Sustained benefit must be evaluated alongside that initial burden and later monitoring requirements.

These observations prevent an easy but unhelpful comparison between the four securities. A longer quoted duration is not inherently superior if it refers to a different endpoint, starts at a different moment or excludes people who never responded. The useful comparison asks whether each program is accumulating the kind of evidence needed for its own clinical proposition. A patient-centered interpretation then connects the statistics to daily function, future treatment needs and the consequences of an event being delayed or avoided.

16Safety has its own observation window

An early safety assessment is strongest for events likely to occur during the period actually observed. Longer follow-up may identify different concerns, including delayed complications or the consequences of persistent biological effects. The absence of a late event before enough time has elapsed cannot establish that the event will never occur.

Sample size matters as much as duration. A small cohort may provide useful information about common acute effects while remaining unable to estimate rare risks precisely. Extending the same small cohort improves the time dimension but does not create the breadth of exposure found in a much larger population. Both dimensions are necessary for a fuller safety picture.

FDA's long-term follow-up guidance for human gene therapy products addresses observation for delayed adverse events according to product characteristics and risk. It should not be read as a claim that every gene-related or cellular product has identical follow-up requirements. Product design, persistence and the specific regulatory context matter.

For lifileucel, the treatment pathway also includes lymphodepletion and interleukin-2 support. FDA's original approval communication identifies serious risks in the regimen and the required clinical setting. Long-term response evidence does not cancel those treatment risks. Conversely, acute adverse events should not be interpreted as proof that a later durable response lacks clinical value. Benefit and burden must be considered together for the authorized population. FDA's Amtagvi announcement.

A careful later update distinguishes new safety findings, persistence of previously observed problems and resolution of acute events. “No new safety signal” is narrower than “no adverse events” and much narrower than “risk-free.” This language matters to investors because safety can affect study conduct, labeling, monitoring, treatment capacity and adoption. It matters even more to patients because a statistical absence in a limited dataset is not an individual guarantee.

17Time cannot supply a missing comparator

A single-arm study can provide compelling observations, particularly when a disease has serious unmet needs and an objective effect is unusual without treatment. Longer follow-up can establish whether those observed effects persist. It does not create a concurrent untreated or alternative-treatment group.

Randomization addresses a different problem: separating the effect of a treatment strategy from differences in the people receiving it and the care around it. A randomized comparison can still face missing data, crossover and limited precision, but it starts with a design intended to make the groups comparable. The passage of time does not make a single-arm cohort acquire that property.

Historical comparisons can provide context without carrying the same evidentiary strength. Patients may differ in disease severity, prior therapies, eligibility, supportive care or assessment frequency. Medical practice can change between the historical period and the current trial. A better outcome in the later study could reflect treatment, those differences or a mixture of them.

The distinction is visible across the four programs. Candel's prostate cancer trial provides a randomized control. DAYBREAK includes an active comparator. KYSA-8 is single-arm. Lifileucel's pivotal melanoma evidence and its subsequent long-term follow-up need to be interpreted within their stated study design. Each can contribute useful knowledge; their percentages should not be placed into an informal league table.

For investment analysis, a stronger design can reduce particular uncertainties, while a striking single-arm effect may justify further development. The question is which uncertainty has actually been reduced. Additional follow-up might make response persistence more credible without resolving comparative effectiveness. A new randomized study might address comparison while still needing time for durability. Treating those as separate advances produces a more accurate understanding of development progress than one generic label such as “de-risked.”

18Landmark analyses and the trap of surviving long enough

A comparison can become biased when groups are defined using something that happens after follow-up starts. For example, patients classified as having a late response must remain alive and observable long enough to enter that category. Comparing their survival from the original starting date with everyone who never reached the category can give the first group an artificial advantage.

This is often described as guarantee-time or immortal-time bias. It is a problem in the analysis, not a statement that any person was literally protected from death. The methodological paper Challenges of Guarantee-Time Bias, Journal of Clinical Oncology, 2013; DOI 10.1200/JCO.2013.49.5283 discusses approaches including landmark analyses and methods that account for changes over time.

A landmark analysis defines a particular time and compares eligible participants from that point onward. It can answer a useful conditional question, such as what happens after a specified visit among those who have reached it. It does not automatically describe outcomes for everyone who originally entered the study. Early events and exclusions need to remain visible.

The same reasoning applies to claims about people who remain off treatment at one year or achieve a complete response after an extended interval. Their subsequent course can be clinically interesting. It cannot be used without qualification to promise the same course to a newly treated person who has not yet reached that state.

For company analysis, this is another reason to ask whether an attractive late result applies to the full population, a predefined subgroup or a selected survivor population. A durable-benefit subgroup may be real and important. Its existence and its frequency are separate questions. Good reporting supplies both. More months of observation can sharpen the conditional story while leaving the original probability of reaching that condition as an equally important part of the treatment's value.

19Waiting for evidence consumes real resources

Clinical observation is not financially free. Sites need support, data must be collected and reviewed, manufacturing capabilities may need to be maintained, and staff continue working between headline events. Companies also fund activities that run ahead of clinical certainty, including regulatory preparation and potential commercial readiness.

Kodiak reported second-quarter 2026 R&D expense of $56.1 million and G&A expense of $10.8 million. Their sum, $66.9 million, provides a simple picture of the reported operating-expense mix for that quarter. The accompanying composition chart divides those two categories. It does not divide cash spending by program and does not measure the cost of waiting specifically for DAYBREAK. Kodiak's quarterly results.

That distinction prevents a common analytical shortcut. R&D expense contains accounting items and work across multiple activities. It cannot be assigned entirely to one near-term catalyst merely because that catalyst dominates attention. The same quarter can fund ongoing trials, manufacturing, employees and development work for several candidates.

Candel's reported R&D expense increased from $7.0 million in the second quarter of 2025 to $19.8 million in the corresponding 2026 quarter. The company linked the increase to clinical, manufacturing and employee-related costs. Its regulatory preparation and potential launch activities demonstrate why the financial clock can accelerate before a product generates sales. The chart comparing Candel and Kodiak shows changes in their own reported expense bases; it does not rank the productivity of their research. Candel's quarterly results.

For an investor, the useful relationship connects the expected information gained with the resources needed to obtain it. A further year might substantially clarify a rare late outcome, or it might add little to the decision that matters next. Understanding the study design helps distinguish those situations. The answer cannot be found by dividing one quarterly expense figure by the number of press releases.

20Cash runway and clinical maturity are different forecasts

A cash balance is a dated amount. A runway statement is management's estimate of how long available resources can support an operating plan. Neither directly measures how mature a clinical dataset is. A well-funded company can still have little long-term evidence, while a mature dataset can belong to a company facing substantial spending before commercialization.

At June 30, 2026, Candel reported $201.6 million in cash and cash equivalents and an operating-plan runway into the first quarter of 2028. Kodiak reported $125.9 million in cash and cash equivalents and expected its current resources to support operations into 2027. These are company-specific forecasts with different operating plans. They should not be converted into precise monthly exhaustion dates from rounded release figures.

Iovance has a commercial revenue stream alongside its clinical obligations. Its second-quarter release reported $99.3 million of total product revenue and a roughly $304 million cash position, whose definition includes cash, cash equivalents, short-term investments and restricted cash. The definition is broader than the cash-and-equivalents balances above. The same release said prior annual revenue guidance was under review; that statement was not itself a newly issued higher range. Iovance's August 6 results.

The economic analysis must also avoid equating net loss with cash consumption. Noncash compensation, working capital, financing flows and other accounting items can make them differ. Even operating cash flow can fluctuate with timing. A forecast based on a single quarter can therefore suggest precision the underlying business does not support.

What matters is whether the funded plan can reach and respond to meaningful evidence. A positive readout may create additional spending needs, including larger studies, validation work or launch preparation. A delay may extend fixed commitments. A negative result may change the plan entirely. Clinical maturity informs these choices; cash availability determines which choices remain feasible. Neither is a substitute for the other.

Source dates: Candel and Kodiak, August 13, 2026; Iovance, August 6, 2026. Runway is management guidance, not a guaranteed exhaustion date. Candel · Kodiak · Iovance.

21Three ways the next interval can change the thesis

The most constructive scenario is that additional observation supports the clinical proposition with better precision. More patients reach a relevant time point, the benefit persists, patient disposition remains clear and no material new safety problem emerges. The result is stronger because it answers an uncertainty that previously mattered, rather than simply because the headline sounds more confident.

A second scenario is that the principal result remains favorable while a commercially important detail becomes less attractive. Examples could include a greater need for additional treatment, a narrower group with sustained benefit or a shorter duration than early observations suggested. These are hypothetical possibilities, not forecasts for the four companies. They explain why a study can meet its main endpoint while the practical treatment proposition still needs careful interpretation.

A third scenario is that the evidence remains difficult to interpret. Missing observations, a small late risk set, changing analysis populations or insufficient events can leave uncertainty unresolved. A later date alone does not solve that problem. The next useful action may be continued follow-up, a different analysis consistent with the protocol or another study designed to answer the remaining question.

For CADL, the key distinction is between the established DFS analysis and less mature later clinical outcomes. For IOVA, it is the relationship between response frequency, durability among responders and outcomes in the wider population. For KOD, it is the combination of scheduled visual outcomes and actual dosing burden. For KYTX, it is persistence of functional improvement with transparent treatment and safety accounting.

The financial consequences depend on the size of the remaining development step. Better evidence can improve the basis for regulatory and commercial planning. It can also reveal that further investment is required. A disciplined interpretation therefore asks two linked questions after every update: what has become more certain, and what still has to happen before that knowledge can translate into a sustainable business?

22A practical way to read the next clinical update

The first pass should identify the population, treatment, comparator, endpoint and starting point of the clock. The second should establish the data cut, analysis set and amount of observation at the time horizon that matters. The third should examine events, missing information, confidence intervals and the distinction between primary and exploratory findings. Only then does the word “durable” acquire a precise meaning.

For a time-to-event result, the curve and number-at-risk table help reveal where evidence is substantial and where it becomes thin. For a scheduled functional assessment, the number with usable measurements and the handling of treatment changes matter. For response duration, the responder denominator must remain visible. For safety, duration and breadth of exposure both constrain what can be concluded.

The underlying sources have different roles. Company releases provide dated operational and financial updates. Peer-reviewed clinical reports provide fuller methods and results. Regulatory documents define the authorized population and explain how evidence is evaluated. Methodological papers help identify errors that recur across diseases. None should be asked to answer a question outside its scope: a quarterly release cannot establish a lifetime clinical outcome, and a promising abstract cannot authorize a product.

The Candel and Iovance company histories provide further context through the Candel Therapeutics Stock Hub and the Iovance Biotherapeutics Stock Hub. The relevant comparison remains evidence over time, with each company's clinical and financial claims attached to their own dates and definitions.

Six months and two years are different because the later interval contains information that did not previously exist. Its value depends on who was observed, what was measured and how uncertainty changed. A credible long-term result does more than extend a timeline: it makes the treatment's actual pattern of benefit and burden easier to understand. That is the progress capable of supporting both better clinical judgments and a more grounded assessment of the business behind the science.