Time-to-Event Data & Censoring - Biostatistics

1. A survival analysis includes 200 subjects followed for up to 10 years. The dataset shows 80 deaths, 60 subjects lost to follow-up, and 60 subjects alive at 10 years. What is the censoring proportion in this study?
   Answer: 60%, because 120 out of 200 subjects were censored When analyzing survival data, you need to understand what censoring means and how to calculate the censoring proportion. Censoring occurs whenever you don't observe the actual event time for a subject - this happens both when subjects are lost to follow-up and when they're still alive at the end of the study period. In this study, censored observations include the 60 subjects lost to follow-up (you don't know when they might have died) plus the 60 subjects still alive at 10 years (you don't know if or when they'll experience the event after study completion). This gives you $60 + 60 = 120$ censored subjects out of 200 total subjects, making the censoring proportion $\frac{120}{200} = 60\%$. Option A incorrectly counts only the subjects lost to follow-up (60/200 = 30%), missing the fact that subjects alive at study end are also censored. Option B confuses the event rate with censoring - the 80 deaths represent observed events, not censored observations. Option D attempts to count all non-deaths (140/200 = 70%), but this incorrectly includes the 80 deaths as if they were censored, when deaths are actually complete observations of the event. Remember that in survival analysis, censoring includes any situation where you don't observe the complete time to event - whether due to loss to follow-up, administrative end of study, or competing risks. Always add up all types of incomplete observations when calculating censoring proportions.
2. A researcher studies time to first heart attack in a cohort study. Some participants move away and cannot be contacted after certain time points. If the probability of moving is related to participants' stress levels (which also affect heart attack risk), but stress levels were not measured in the study, what analytical consideration is most important?
   Answer: Sensitivity analyses should assess potential bias from informative censoring due to unmeasured confounders When analyzing survival data, you must carefully consider whether censoring is "informative" - meaning the reason for censoring is related to the outcome of interest. This scenario presents a classic case of potentially informative censoring that threatens the validity of standard survival analysis. The correct approach is C: conducting sensitivity analyses to assess potential bias from informative censoring. Since participants who move away do so partly because of stress levels (which also affect heart attack risk), their censoring isn't random - it's systematically related to the unmeasured confounder (stress) that influences the outcome. This violates the "independent censoring" assumption required for valid survival analysis. Sensitivity analyses can explore how different assumptions about the relationship between stress, moving, and heart attack risk might bias your results. A is incorrect because censoring is NOT independent when moving probability depends on unmeasured stress levels that affect the outcome. The independence assumption is violated even if censoring appears independent of measured covariates. B is wrong because you cannot impute data for participants who have left the study - you have no information about their stress levels or subsequent outcomes to base imputations on. D misapplies competing risk methodology. Moving isn't a competing risk that prevents heart attacks from occurring; participants could still have heart attacks after moving. The issue is loss of information, not competing events. Study tip: In survival analysis questions, always ask: "Is the reason for censoring related to the outcome?" If yes (especially through unmeasured variables), consider informative censoring and the need for sensitivity analyses to assess potential bias.
3. A study follows patients for development of diabetes over 5 years. Patients with family history of diabetes are more likely to drop out of the study due to anxiety about their risk, and these patients also have higher actual risk of developing diabetes. What is the most concerning implication for survival analysis?
   Answer: Underestimation of diabetes incidence due to loss of high-risk individuals When analyzing longitudinal studies like this diabetes follow-up, you need to recognize how dropout patterns can create systematic bias in your results. The key issue here is informative censoring — when the reason people leave the study is related to the outcome you're measuring. In this scenario, patients with family history are both more likely to drop out AND more likely to develop diabetes. When these high-risk individuals systematically leave the study, you're left with a sample that's no longer representative of your original population. This creates a serious underestimation of the true diabetes incidence rate, making option C correct. Let's examine why the other options miss the mark: A focuses on statistical power, which is about your ability to detect effects, not about bias in your estimates. While fewer participants does reduce power, the bigger problem is that your remaining sample gives you wrong answers, not just less precise ones. B mentions proportional hazards violation, but the dropout pattern described doesn't inherently violate the assumption that hazard ratios remain constant over time. D suggests competing risks analysis, but anxiety-related dropout isn't a competing outcome that prevents diabetes — it's just censoring. The core principle to remember: in survival analysis, always ask whether censoring is random or informative. When people at higher risk systematically drop out, your survival curves will be overly optimistic, underestimating the true incidence of your outcome. This type of bias can completely invalidate your study conclusions.
4. A study of time to cancer recurrence follows patients for 10 years post-surgery. Patients undergo imaging every 3 months for the first 2 years, every 6 months for years 3-5, and annually thereafter. A patient has no evidence of recurrence at the 18-month scan, misses the 21-month scan, and is found to have recurrence at the 2-year scan. The study protocol specifies that recurrence time should be estimated as the midpoint between the last negative and first positive scan. What potential bias does this approach introduce?
   Answer: Measurement bias due to varying imaging intervals, with systematic underestimation of recurrence times in later follow-up periods The midpoint imputation method will systematically underestimate recurrence times more in later follow-up periods when imaging intervals are longer (6 months or 1 year vs. 3 months early on). This creates differential measurement bias across follow-up time. A is wrong because more frequent early monitoring would lead to more accurate, not biased, early detection. B describes a different issue (missed visits) but doesn't address the systematic bias from varying intervals. C incorrectly applies length bias, which relates to prevalent vs. incident cases.
5. A clinical trial comparing two treatments for heart failure uses a composite endpoint of "time to cardiovascular death or heart transplantation." During the study, 25 patients died from cardiovascular causes, 8 patients received heart transplants, 12 patients died from non-cardiovascular causes, and 45 patients completed follow-up alive without transplant. How should the 12 patients who died from non-cardiovascular causes be handled in the survival analysis?
   Answer: Analyzed using competing risks methods since non-cardiovascular death prevents observation of the primary endpoint Non-cardiovascular death is a competing risk that prevents observation of the primary composite endpoint (CV death or transplant). Standard survival analysis assumes patients could eventually experience the event if followed long enough, but competing risks make this impossible. A is wrong because this isn't simple censoring. B is wrong because these patients provide information about treatment effects. D is wrong because the endpoint is specifically defined as CV death or transplant, not all-cause death.
6. A study examines time to viral rebound in HIV patients after stopping treatment. The study protocol mandates that follow-up stops immediately when a patient's viral load becomes undetectable (indicating successful alternative treatment), even if rebound might occur later. For the outcome 'time to viral rebound,' how should patients with undetectable viral loads be classified?
   Answer: Right-censored observations because rebound could still occur after viral suppression When analyzing survival data, you need to distinguish between different types of incomplete observations. Censoring occurs when you know the event of interest (viral rebound) hasn't happened by the observation time, but you don't know if or when it will occur after that point. In this scenario, patients with undetectable viral loads represent right-censored observations because viral rebound could theoretically still occur after the study ends. The key insight is that "undetectable" doesn't mean "impossible" - it means the event hasn't been observed yet. These patients contribute valuable information about survival up to their last observation point, but we don't know their ultimate outcome. Option A is incorrect because undetectable viral load doesn't represent the actual event of interest (rebound) - it's simply the absence of the event at that time point. Option C misapplies competing risks, which would apply if viral suppression permanently prevented rebound from ever occurring. However, suppression is potentially temporary, so rebound remains possible. Option D suggests informative censoring, but the protocol-mandated stopping rule based on viral suppression doesn't necessarily create bias related to rebound risk - it's an administrative decision rather than one driven by prognosis. The distinction here is crucial: temporary viral suppression with protocol-mandated follow-up cessation creates right-censoring, not competing risks or exact observations. Study tip: In survival analysis, ask yourself "Could the event still happen later?" If yes, it's likely censored. If the competing event permanently prevents your outcome of interest, then consider competing risks.
7. In a clinical trial comparing two treatments for cancer, the study ends when the last enrolled patient completes 3 years of follow-up. Patients enrolled earlier have longer potential follow-up times than those enrolled later. What type of censoring occurs for patients who don't experience events by study end?
   Answer: Type I administrative censoring because all patients are censored at the same calendar date When analyzing censoring in survival studies, focus on what determines the censoring time and whether it's planned or random. This question describes a common clinical trial design where enrollment occurs over time but all patients are followed until the same predetermined end date. The correct answer is A because this represents Type I administrative censoring. In this scenario, the study has a fixed calendar end date when the last enrolled patient reaches 3 years of follow-up. All patients who haven't experienced the event by this predetermined date are censored at the same time point. The key characteristic of Type I censoring is that censoring occurs at a pre-specified time, regardless of how many events have been observed. Option B is incorrect because Type II administrative censoring occurs when you stop the study after observing a fixed number of events (like stopping after 50 deaths), not after a predetermined time period. Option C misses the mark because this isn't random censoring - the censoring time is planned and administrative, not due to random factors like patients moving away or dropping out for unrelated reasons. Option D confuses the enrollment pattern with the censoring mechanism. While patients do enter at different times (staggered entry), the question specifically asks about the censoring type for patients who don't experience events by study end. Remember: Type I censoring = fixed time endpoint, Type II censoring = fixed number of events. Don't let staggered enrollment distract you from identifying the actual censoring mechanism at study termination.
8. In a study of time to kidney transplant, some patients on the waiting list die before receiving a transplant, while others receive transplants or are still waiting when the study ends. For the specific outcome 'time to transplant,' how should deaths before transplant be handled?
   Answer: Treat as right-censored observations at the time of death When analyzing survival data, you need to distinguish between different types of incomplete observations. In this kidney transplant study, you're specifically measuring "time to transplant," making transplantation your event of interest, not death. Patients who die before receiving a transplant represent right-censored observations. These patients were observed until their death, but we never observed the event we're studying (transplantation). Right-censoring occurs when follow-up ends before the event happens, whether due to death, loss to follow-up, or study completion. The key insight is that these patients could have eventually received a transplant if they had lived longer—we simply don't know when. Therefore, option C correctly treats deaths as right-censored at the time of death. Option A is wrong because excluding these patients would create bias and waste valuable information about waiting times. Option B incorrectly treats death as if it were transplantation, which fundamentally misrepresents what happened to these patients. Option D might seem tempting because death and transplantation are indeed competing events, but the question specifically asks about analyzing "time to transplant" as a single outcome. While competing risks analysis could be appropriate for a broader study, standard survival analysis with right-censoring directly answers the question posed. Study tip: Remember that censoring depends on your specific research question. The same patients could be events in one analysis (studying time to death) but censored observations in another (studying time to transplant). Always identify your event of interest first.
9. In a clinical trial studying time to disease recurrence, 200 patients were followed for 5 years. At the end of the study, 45 patients had experienced recurrence, 120 patients completed the full 5-year follow-up without recurrence, and 35 patients were lost to follow-up at various times. If the hazard rate is constant over time, what is the most appropriate interpretation of the censoring pattern in this study?
   Answer: The censoring is non-informative assuming loss to follow-up is independent of recurrence risk, and both administrative and random censoring are present This study has both types of censoring: administrative censoring (120 patients censored at 5 years) and random censoring (35 patients lost to follow-up). Under the assumption that loss to follow-up is independent of recurrence risk, the censoring is non-informative. A is wrong because it assumes informative censoring without justification. C is wrong because it ignores the 35 patients lost to follow-up. D is wrong because we can characterize the censoring mechanism from the given information.
10. In a study of time to viral suppression in HIV patients, researchers find that patients with higher baseline viral loads are more likely to discontinue treatment early due to side effects. This early discontinuation prevents observation of suppression times. What is the primary concern for survival analysis?
    Answer: Informative censoring may bias estimates because censoring depends on suppression likelihood When you encounter survival analysis questions, focus on the key assumption that censoring must be non-informative—meaning the reason for censoring shouldn't be related to the outcome of interest. In this HIV study, patients with higher baseline viral loads are more likely to discontinue treatment due to side effects. This creates a serious problem: the patients who drop out (get censored) are systematically different from those who continue. Specifically, they're the patients who likely would have taken longer to achieve viral suppression or might never have achieved it. When censoring is related to the probability of experiencing the event, it's called informative censoring, and it violates survival analysis assumptions. This will bias time-to-suppression estimates because you're systematically losing the "harder to treat" patients from your analysis. Answer C correctly identifies this as informative censoring bias. Answer A is wrong because sample size affects power but doesn't address the fundamental bias issue here. Answer B incorrectly suggests multiple imputation, but you can't impute unknown survival times—survival analysis is specifically designed to handle incomplete follow-up through censoring methods. Answer D misidentifies this as administrative censoring, which occurs when studies end at predetermined times regardless of patient characteristics, creating non-informative censoring. Remember this pattern: whenever dropout or censoring is related to patient characteristics that also affect the outcome probability, suspect informative censoring bias. This is one of the most important violations of survival analysis assumptions to recognize.
11. A multi-center study of time to viral suppression in HIV patients uses different laboratory assays across sites. Site A uses an assay with a lower limit of detection of 50 copies/mL, while Site B uses an assay detecting down to 20 copies/mL. Both sites define viral suppression as "undetectable" by their respective assays. When combining data for analysis, what issue does this create for time-to-event analysis?
    Answer: Information bias due to varying detection limits, where Site A may misclassify patients with 20-49 copies/mL as having achieved suppression Site A's higher detection limit (50 copies/mL) means patients with viral loads between 20-49 copies/mL will be classified as 'undetectable' (suppressed) when they might not meet the more stringent criterion used at Site B. This creates information bias where the same biological state is classified differently. A is backwards - Site A would appear to achieve suppression earlier, not later. C addresses a different type of bias not directly related to assay differences. D incorrectly applies competing risks concepts to what is essentially a measurement issue.
12. In a survival study of cancer patients, some participants moved to different countries and could not be contacted after certain time points. If the probability of moving depends on the patient's unobserved health status at the time of moving, what type of censoring mechanism is this?
    Answer: Informative censoring that may bias survival estimates if not properly accounted for When analyzing censoring mechanisms in survival studies, you need to determine whether the reason for censoring is related to the outcome being studied. This distinction is crucial because different types of censoring affect the validity of your survival estimates differently. In this scenario, patients are lost to follow-up because they moved countries, and crucially, the probability of moving depends on their unobserved health status. This creates informative censoring (also called non-ignorable censoring), where the censoring mechanism is directly related to the survival outcome. For example, sicker patients might be more likely to move back to their home countries for family support, while healthier patients might move for career opportunities. Since we can't observe their health status at the time of moving, we can't account for this relationship, making option C correct. Option A is wrong because MCAR censoring occurs when the probability of being censored is completely independent of both observed and unobserved patient characteristics. Option B is incorrect because MAR censoring means the censoring depends only on observed covariates, not unobserved health status. While you can adjust for MAR censoring using statistical techniques, you cannot adjust for the unobserved health status described here. Option D is wrong because administrative censoring occurs at predetermined times (like study end dates) and is independent of patient characteristics. Study tip: Remember the hierarchy of censoring assumptions. MCAR is the least problematic, MAR can be handled with proper adjustment, but informative censoring requires specialized methods or may lead to biased estimates if ignored.
13. A pharmaceutical company studies time to drug approval by regulatory agencies. They begin observation when companies submit applications, but some applications were submitted before the study period began and are already under review. For these applications, what type of censoring issue arises?
    Answer: Left truncation because we only observe applications surviving to the study start When analyzing survival data, you must distinguish between censoring (incomplete observation of the event) and truncation (selective sampling based on survival). This question tests your understanding of left truncation, a common issue in observational studies. Left truncation occurs when you can only observe subjects who have "survived" to enter your study period. Here, applications submitted before the study began represent left truncation because only those applications that hadn't been approved or rejected by the study start date could be observed. Applications that were quickly approved or rejected before the study began are completely missing from your dataset—you never see them at all. Option A is incorrect because right censoring refers to incomplete observation of the endpoint (not knowing when approval occurs), but the issue here is about which applications you can observe in the first place. Option B misses the mark because interval censoring involves knowing an event occurred within a time window but not the exact time—this isn't about imprecise timing of approval. Option D is wrong because administrative censoring refers to ending observation at a predetermined study cutoff time, which isn't the issue described. The key distinction to remember: censoring means you observe the subject but don't see their complete outcome, while truncation means certain subjects never enter your study at all based on their survival status. Left truncation creates bias because your sample excludes cases with very short or very long times to the event that occurred before observation began.
14. In a study of time to return to work after injury, researchers contact participants monthly by phone. A participant reports being at work during the month 4 call but was not working during the month 3 call. For the outcome 'time to return to work,' what additional information would change this from interval censoring to exact observation?
    Answer: Information about the specific date when the participant returned to work When you encounter survival analysis questions involving time-to-event data, the key distinction is understanding different types of censoring and what information transforms one type into another. In this scenario, you know the participant wasn't working at month 3 but was working at month 4. This creates interval censoring because the actual return-to-work event occurred somewhere within that one-month interval between calls. You don't know the exact timing, only that it happened within a specific time window. Option B is correct because knowing the specific date when the participant returned to work would eliminate the uncertainty about timing. Instead of knowing the event occurred "sometime between month 3 and month 4," you'd have the exact moment it occurred, making this a complete (uncensored) observation. Option A is wrong because injury severity and expected recovery time are baseline characteristics that don't provide information about the actual timing of the return-to-work event. Option C is incorrect because work schedule and hours worked describe the nature of the work return but don't pinpoint when it occurred within that month interval. Option D is wrong because motivation is a psychological factor that might influence recovery speed but doesn't give you the precise timing needed to eliminate interval censoring. Study tip: Remember that censoring types are defined by what you know about event timing. Right censoring = you know when follow-up ended but not when the event occurred. Interval censoring = you know the event occurred within a time window. Exact observation = you know precisely when the event occurred.
15. In a study of time to treatment failure in tuberculosis patients, some patients stop taking medication due to side effects before failure can be assessed. These patients have a different risk profile for treatment failure compared to those who continue medication. When analyzing time to treatment failure, what is the most appropriate way to handle patients who stop medication?
    Answer: Treat them as right-censored at medication discontinuation time with potential bias assessment When analyzing time-to-event data like treatment failure in tuberculosis patients, you need to distinguish between different types of incomplete observations. The key concept here is understanding censoring versus competing risks in survival analysis. Patients who discontinue medication due to side effects represent informative censoring - their withdrawal is related to factors that may also influence their risk of treatment failure. Since you cannot observe their treatment failure after discontinuation (they're no longer on treatment), they should be treated as right-censored at the time they stopped medication. However, because this censoring is potentially informative rather than random, you must assess and address the potential bias this creates in your analysis. Option A is incorrect because excluding these patients wastes valuable information and can introduce selection bias. You observed them for a period of time before discontinuation, which contributes to understanding the survival function. Option B is wrong because stopping medication is not the same as treatment failure. These are different events with different clinical meanings - discontinuation due to side effects doesn't indicate the treatment failed to work. Option D incorrectly frames this as a competing risks problem. Competing risks apply when multiple mutually exclusive events can occur (like death from different causes). Here, medication discontinuation and treatment failure aren't truly competing - discontinuation prevents observation of failure rather than competing with it. Study tip: In survival analysis questions, always ask: "Can I still observe my outcome of interest after this event?" If no, it's censoring. If the event itself prevents the outcome, consider whether it's truly competing risk or simply censoring.
16. In a study of time to first hospitalization, researchers used a Cox proportional hazards model. During analysis, they discovered that 15% of patients had experienced their first hospitalization before study enrollment but were still included in the analysis with time zero set at enrollment. What type of bias does this introduce, and how should it be addressed?
    Answer: Length bias due to prevalent cases; these patients should be analyzed separately or the study should focus only on incident cases Including patients who already experienced the event creates length bias because these prevalent cases may have different characteristics than incident cases. The study should ideally focus on incident cases only. A is wrong because this isn't left truncation (we know about these patients). B is wrong because this isn't left censoring (we know the event occurred). D is wrong because immortal time bias refers to periods where events cannot occur, and resetting time zero doesn't make sense for this endpoint.
17. A pharmaceutical company conducts a clinical trial where patients are randomized to drug A or placebo. Due to safety concerns identified during the trial, the sponsor terminates the study early and all remaining patients are switched to drug A. For survival analysis purposes, how should patients originally assigned to placebo be handled from the time of treatment switching?
    Answer: Continue analyzing them in the placebo group using intention-to-treat principles, censoring at the time of switch For regulatory purposes and to preserve randomization, intention-to-treat analysis is preferred, which means censoring placebo patients at the time of switch. This maintains the randomized comparison while acknowledging that we can no longer observe the placebo effect after switching. B violates randomization. C is complex and may not be appropriate for primary analysis. D reduces power and may introduce selection bias.
18. In a study of time to recovery from depression, patients are assessed monthly using standardized scales. A patient shows severe depression at month 2, mild depression at month 4, and full recovery at month 6. For the outcome 'time to any improvement' (severe to mild or better), what observation type applies?
    Answer: Interval censoring between months 2 and 4 when improvement occurred When analyzing time-to-event data with periodic assessments, you need to distinguish between different types of censoring based on what you know about when the event actually occurred versus when it was observed. In this scenario, you know the patient had severe depression at month 2 and mild depression (improvement) at month 4. Since assessments only happen monthly, the actual improvement occurred sometime between these two time points, but you don't know exactly when. This creates interval censoring - you know the event happened within a specific time interval (between months 2 and 4) but not the precise timing. Answer B correctly identifies this situation. Answer A is wrong because you don't have an exact observation - you only know improvement was detected at month 4, not that it occurred exactly then. The improvement could have happened at any point during those two months. Answer C misunderstands the outcome of interest. You're measuring "time to any improvement," and that event has already occurred by month 4. Right censoring would apply if you were measuring time to complete recovery and the study ended before full recovery. Answer D incorrectly suggests left censoring, which occurs when you don't know if the event happened before the study began. Here, you have a clear baseline (severe depression at month 2) before the improvement was detected. Study tip: For time-to-event questions, always identify exactly what event you're measuring and when you know it occurred versus when it was observed. Interval censoring is common in studies with periodic follow-up visits.
19. A researcher analyzes time to employment after job training. Some participants find jobs, others are still searching when the study ends, and some stop searching and are lost to follow-up. If participants who stop searching do so because they become discouraged by lack of progress, how should this censoring be characterized?
    Answer: Informative censoring that may overestimate employment rates if not addressed When analyzing survival data, you must carefully consider why observations are censored, as this directly affects the validity of your results. Censoring occurs when you don't observe the complete time to the event of interest. The key insight here is recognizing informative censoring - when the reason for censoring is related to the outcome being studied. In this scenario, participants stop searching because they're discouraged by their lack of progress in finding employment. This creates a systematic bias because those who drop out are likely the ones who would have taken longer to find jobs (or might never have found them). When you exclude these participants from your analysis, you're left with a sample enriched with more successful job seekers, leading to overestimated employment rates. Option A is wrong because this censoring cannot be ignored - it's directly related to the outcome. Option B incorrectly characterizes this as random censoring, but the decision to stop searching is specifically tied to poor job prospects, not random timing. Option C describes administrative censoring (like ending a study at a predetermined date), but here participants are making personal decisions based on their experiences. Answer: D correctly identifies this as informative censoring that biases results toward better outcomes. Study tip: Always ask yourself "Why are observations censored?" If the censoring mechanism is related to the outcome (like dropping out due to treatment side effects or poor progress), it's informative and requires special analytical approaches. Random or administrative censoring can be handled with standard survival methods.
20. In a longitudinal study of cognitive decline, researchers test participants every 6 months. A participant scores normally at month 12, shows mild impairment at month 18, and has severe impairment at month 24. For the outcome 'time to mild cognitive impairment,' what type of observation is this?
    Answer: Interval censoring between months 12 and 18 when impairment developed When analyzing time-to-event data in survival analysis, you need to identify what type of observation you're dealing with based on when the event actually occurred versus when it was detected. In this scenario, the participant was normal at month 12 and showed mild impairment at month 18. Since participants are only tested every 6 months, we know the mild cognitive impairment developed sometime during that 6-month window, but we don't know the exact moment it occurred. This creates an interval where we know the event happened, making it interval censoring between months 12 and 18 (answer D). Answer A is incorrect because we don't have an exact observation - we only know impairment was detected at month 18, not that it started precisely then. The impairment could have begun days or months earlier within that interval. Answer B misapplies right censoring, which occurs when we lose follow-up on a participant before observing the event of interest. Here, we did observe the event (mild impairment), so there's no censoring of future progression for this particular outcome. Answer C incorrectly suggests left censoring, which would mean the participant already had mild impairment when the study began. Since the participant tested normal at month 12, we know impairment hadn't started by that point. Study tip: In survival analysis, always ask yourself: "Do I know exactly when the event occurred?" If you only know it happened between two observation points, it's interval censoring - a common scenario in studies with scheduled follow-up visits.

Start this quiz