Occupancy in dynamic systems: accounting for multiple scales and false positives using environmental DNA to inform monitoring

Occupancy is an important metric to understand current and future trends in populations that have declined globally. In addition, occupancy can be an efficient tool for conducting landscape-scale and long-term monitoring. A challenge for occupancy monitoring programs is to determine the appropriate spatial scale of analysis and to obtain precise occupancy estimates for elusive species. We used a multi-scale occupancy model to assess occupancy of Columbia spotted frogs in the Great Basin, USA, based on environmental DNA (eDNA) detections. We collected three replicate eDNA samples at 220 sites across the Great Basin. We estimated and modeled ecological factors that described watershed and site occupancy at multiple spatial scales simultaneously while accounting for imperfect detection. Additionally, we conducted visual and dipnet surveys at all sites and used our paired detections to estimate the probability of a false positive detection for our eDNA sampling. We applied the estimated false positive rate to our multi-scale occupancy dataset and assessed changes in model selection. We had higher naïve occupancy estimates for eDNA (0.37) than for traditional survey methods (0.20). We estimated our false positive detection rate per qPCR replicate at 0.023 (95% CI: 0.016–0.033). When the false positive rate was applied to the multi-scale dataset, we did not observe substantial changes in model selection or parameter estimates. Conservation and resource managers have an increasing need to understand species occupancy in highly variable landscapes where the spatial distribution of habitat changes significantly over time due to climate change and human impact. A multi-scale occupancy approach can be used to obtain regional occupancy estimates that can account for spatially dynamic differences in availability over time, especially when assessing potential declines. Additionally, this study demonstrates how eDNA can be used as an effective tool for improved occupancy estimates across broad geographic scales for long-term monitoring.     

Experimental Investigation of Observation Error in Anuran Call Surveys

ABSTRACT Occupancy models that account for imperfect detection are often used to monitor anuran and songbird species occurrence. However, presence—absence data arising from auditory detections may be more prone to observation error (e.g., false-positive detections) than are sampling approaches utilizing physical captures or sightings of individuals. We conducted realistic, replicated field experiments using a remote broadcasting system to simulate simple anuran call surveys and to investigate potential factors affecting observation error in these studies. Distance, time, ambient noise, and observer abilities were the most important factors explaining false-negative detections. Distance and observer ability were the best overall predictors of false-positive errors, but ambient noise and competing species also affected error rates for some species. False-positive errors made up 5% of all positive detections, with individual observers exhibiting false-positive rates between 0.5% and 14%. Previous research suggests false-positive errors of these magnitudes would induce substantial positive biases in standard estimators of species occurrence, and we recommend practices to mitigate for false positives when developing occupancy monitoring protocols that rely on auditory detections. These recommendations include additional observer training, limiting the number of target species, and establishing distance and ambient noise thresholds during surveys.     

Replication levels,false presences and the estimation of the presence/absence from eDNA metabarcoding data

Environmental DNA (eDNA) metabarcoding is increasingly used to study the present and past biodiversity. eDNA analyses often rely on amplification of very small quantities or degraded DNA. To avoid missing detection of taxa that are actually present (false negatives), multiple extractions and amplifications of the same samples are often performed. However, the level of replication needed for reliable estimates of the presence/absence patterns remains an unaddressed topic. Furthermore, degraded DNA and PCR/sequencing errors might produce false positives. We used simulations and empirical data to evaluate the level of replication required for accurate detection of targeted taxa in different contexts and to assess the performance of methods used to reduce the risk of false detections. Furthermore, we evaluated whether statistical approaches developed to estimate occupancy in the presence of observational errors can successfully estimate true prevalence, detection probability and false‐positive rates. Replications reduced the rate of false negatives; the optimal level of replication was strongly dependent on the detection probability of taxa. Occupancy models successfully estimated true prevalence, detection probability and false‐positive rates, but their performance increased with the number of replicates. At least eight PCR replicates should be performed if detection probability is not high, such as in ancient DNA studies. Multiple DNA extractions from the same sample yielded consistent results; in some cases, collecting multiple samples from the same locality allowed detecting more species. The optimal level of replication for accurate species detection strongly varies among studies and could be explicitly estimated to improve the reliability of results.     

A model‐based solution for observational errors in laboratory studies

Molecular techniques for detecting microorganisms, macroorganisms and infectious agents are susceptible to false‐negative and false‐positive errors. If left unaddressed, these observational errors may yield misleading inference concerning occurrence, prevalence, sensitivity, specificity and covariate relationships. Occupancy models are widely used to account for false‐negative errors and more recently have even been used to address false‐positive errors, too. Current modelling options assume false‐positive errors only occur in truly negative samples, an assumption that yields biased inference concerning detection because a positive sample could be classified as such not because the target agent was successfully detected, but rather due to a false‐positive test result. We present an extension to the occupancy modelling framework that allows false‐positive errors in both negative and positive samples, thereby providing unbiased inference concerning occurrence and detection, as well as reliable conclusions about the efficacy of sampling designs, handling protocols and diagnostic tests. We apply the model to simulated data, showing that it recovers known parameters and outperforms other approaches that are commonly used when confronted with observation errors. We then apply the model to an experimental data set on Batrachochytrium dendrobatidis, a pathogenic fungus that is implicated in the global decline or extinction of hundreds of amphibian species. The model‐based approach we present is not only useful for obtaining reliable inference when data are contaminated with observational errors, but also eliminates the need for establishing arbitrary thresholds or decision rules that have hidden and unintended consequences.     

Conditionally autoregressive models improve occupancy analyses of autocorrelated data: An example with environmental DNA

Site occupancy‐detection models (SODMs) are statistical models widely used for biodiversity surveys where imperfect detection of species occurs. For instance, SODMs are increasingly used to analyse environmental DNA (eDNA) data, taking into account the occurrence of both false‐positive and false‐negative errors. However, species occurrence data are often characterized by spatial and temporal autocorrelation, which might challenge the use of standard SODMs. Here we reviewed the literature of eDNA biodiversity surveys and found that most of studies do not take into account spatial or temporal autocorrelation. We then demonstrated how the analysis of data with spatial or temporal autocorrelation can be improved by using a conditionally autoregressive SODM, and show its application to environmental DNA data. We tested the autoregressive model on both simulated and real data sets, including chronosequences with different degrees of autocorrelation, and a spatial data set on a virtual landscape. Analyses of simulated data showed that autoregressive SODMs perform better than traditional SODMs in the estimation of key parameters such as true‐/false‐positive rates and show a better discrimination capacity (e.g., higher true skill statistics). The usefulness of autoregressive SODMs was particularly high in data sets with strong autocorrelation. When applied to real eDNA data sets (eDNA from lake sediment cores and freshwater), autoregressive SODM provided more precise estimation of true‐/false‐positive rates, resulting in more reasonable inference of occupancy states. Our results suggest that analyses of occurrence data, such as many applications of eDNA, can be largely improved by applying conditionally autoregressive specifications to SODMs.     

Identifying error and accurately interpreting environmental DNA metabarcoding results: A case study to detect vertebrates at arid zone waterholes

Environmental DNA (eDNA) metabarcoding surveys enable rapid, noninvasive identification of taxa from trace samples with wide‐ranging applications from characterizing local biodiversity to identifying food‐web interactions. However, the technique is prone to error from two major sources: (a) contamination through foreign DNA entering the workflow, and (b) misidentification of DNA within the workflow. Both types of error have the potential to obscure true taxon presence or to increase taxonomic richness by incorrectly identifying taxa as present at sample sites, but multiple error sources can remain unaccounted for in metabarcoding studies. Here, we use data from an eDNA metabarcoding study designed to detect vertebrate species at waterholes in Australia's arid zone to illustrate where and how in the workflow errors can arise, and how to mitigate those errors. We detected the DNA of 36 taxa spanning 34 families, 19 orders and five vertebrate classes in water samples from waterholes, demonstrating the potential for eDNA metabarcoding surveys to provide rapid, noninvasive detection in remote locations, and to widely sample taxonomic diversity from aquatic through to terrestrial taxa. However, we initially identified 152 taxa in the samples, meaning there were many false positive detections. We identified the sources of these errors, allowing us to design a stepwise process to detect and remove error, and provide a template to minimize similar errors that are likely to arise in other metabarcoding studies. Our findings suggest eDNA metabarcoding surveys need to be carefully conducted and screened for errors to ensure their accuracy.     

The consequences of using indirect signs that decay to determine species' occupancy

Statistical models of species' distributions rely on data on species' occupancy, or use, of sites across space and/or time. For rare or cryptic species, indirect signs, such as dung, may be the only realistic means of determining their occupancy status across broad spatial extents. However, the consequences of sign decay for errors in estimates of occupancy have not previously been considered. If signs decay very rapidly, then false‐negative errors may occur because signs at an occupied site have decayed by the time it is surveyed. On the other hand, if signs decay very slowly, false‐positive errors may occur because signs remain present at sites that are no longer occupied. We addressed this issue by quantifying, as functions of sign decay and accumulation rates: 1) the false‐negative error rate due to sign decay and, 2) the expected time interval prior to a survey within which signs indicate the species was present; as this time interval increases, false‐positives become more likely. We then applied this to the specific example of koala Phascolarctos cinereus occupancy derived from faecal pellet surveys using data on faecal pellet decay rates. We show that there is a clear trade‐off between false‐negative error rates and the potential for false‐positive errors. For the koala case study, false‐negative errors were low on average and the expected time interval prior to surveys that detected pellets indicate the species was present within less than 2–3 yr. However, these quantities showed quite substantial spatial variation that could lead to biased parameter estimates for distribution models based on faecal pellet surveys. This highlights the importance of observation errors arising from sign decay and we suggest some modifications to existing methods to deal with this issue.     

The importance of including imperfect detection models in eDNA experimental design

Environmental DNA (eDNA) is DNA that has been isolated from field samples, and it is increasingly used to infer the presence or absence of particular species in an ecosystem. However, the combination of sampling procedures and subsequent molecular amplification of eDNA can lead to spurious results. As such, it is imperative that eDNA studies include a statistical framework for interpreting eDNA presence/absence data. We reviewed published literature for studies that utilized eDNA where the species density was known and compared the probability of detecting the focal species to the sampling and analysis protocols. Although biomass of the target species and the volume per sample did not impact detectability, the number of field replicates and number of samples from each replicate were positively related to detection. Additionally, increased number of PCR replicates and increased primer specificity significantly increased detectability. Accordingly, we advocate for increased use of occupancy modelling as a method to incorporate effects of sampling effort and PCR sensitivity in eDNA study design. Based on simulation results and the hierarchical nature of occupancy models, we suggest that field replicates, as opposed to molecular replicates, result in better detection probabilities of target species.     

ednaoccupancy: An r package for multiscale occupancy modelling of environmental DNA data

In this article, we describe ednaoccupancy , an r package for fitting Bayesian, multiscale occupancy models. These models are appropriate for occupancy surveys that include three nested levels of sampling: primary sample units within a study area, secondary sample units collected from each primary unit and replicates of each secondary sample unit. This design is commonly used in occupancy surveys of environmental DNA (eDNA). ednaoccupancy allows users to specify and fit multiscale occupancy models with or without covariates, to estimate posterior summaries of occurrence and detection probabilities, and to compare different models using Bayesian model‐selection criteria. We illustrate these features by analysing two published data sets: eDNA surveys of a fungal pathogen of amphibians and eDNA surveys of an endangered fish species.     

Assessing Environmental DNA Detection in Controlled Lentic Systems

Little consideration has been given to environmental DNA (eDNA) sampling strategies for rare species. The certainty of species detection relies on understanding false positive and false negative error rates. We used artificial ponds together with logistic regression models to assess the detection of African jewelfish eDNA at varying fish densities (0, 0.32, 1.75, and 5.25 fish/m³). Our objectives were to determine the most effective water stratum for eDNA detection, estimate true and false positive eDNA detection rates, and assess the number of water samples necessary to minimize the risk of false negatives. There were 28 eDNA detections in 324, 1-L, water samples collected from four experimental ponds. The best-approximating model indicated that the per-L-sample probability of eDNA detection was 4.86 times more likely for every 2.53 fish/m³ (1 SD) increase in fish density and 1.67 times less likely for every 1.02 C (1 SD) increase in water temperature. The best section of the water column to detect eDNA was the surface and to a lesser extent the bottom. Although no false positives were detected, the estimated likely number of false positives in samples from ponds that contained fish averaged 3.62. At high densities of African jewelfish, 3–5 L of water provided a >95% probability for the presence/absence of its eDNA. Conversely, at moderate and low densities, the number of water samples necessary to achieve a >95% probability of eDNA detection approximated 42–73 and >100 L, respectively. Potential biases associated with incomplete detection of eDNA could be alleviated via formal estimation of eDNA detection probabilities under an occupancy modeling framework; alternatively, the filtration of hundreds of liters of water may be required to achieve a high (e.g., 95%) level of certainty that African jewelfish eDNA will be detected at low densities (i.e., <0.32 fish/m³ or 1.75 g/m³).     

Improved environmental DNA sampling scheme for Alabama sturgeon provides new insight into a species once presumed extinct

Detection of rare species can be challenging and time-consuming using conventional methods, but environmental DNA (eDNA) is becoming a commonly used tool for detection in conservation and management of species. This study demonstrates the utility of the precipitation method (precipitated and preserved in 3 M sodium acetate and 95% ethanol) for collection of eDNA to detect the seasonal distribution of the critically endangered Alabama sturgeon (Scaphirhynchus suttkusi). Surface and benthic water samples were collected across a wider geographic area than previously published for Alabama sturgeon eDNA. Surface and benthic samples both yielded detections and resulted in a similar proportion of positive detections to previous work. However, by sampling a greater portion of the distribution of the Alabama sturgeon, further insight was provided on potential sturgeon movement. The results of the precipitation method show that Alabama sturgeon detections increase during spawning months, and that the fish may be overwintering in the Tombigbee River. High detections from winter benthic samples suggest that habitat choice may play a role in detectability and highlight the need to consider natural history when designing environmental DNA studies. When designing environmental DNA collection for rare species, sampling design should factor in species ecology, habitat use, site characteristics, and specific questions driving the research.     

Improving geographically extensive acoustic survey designs for modeling species occurrence with imperfect detection and misidentification

Acoustic recording units (ARUs) enable geographically extensive surveys of sensitive and elusive species. However, a hidden cost of using ARU data for modeling species occupancy is that prohibitive amounts of human verification may be required to correct species identifications made from automated software. Bat acoustic studies exemplify this challenge because large volumes of echolocation calls could be recorded and automatically classified to species. The standard occupancy model requires aggregating verified recordings to construct confirmed detection/non‐detection datasets. The multistep data processing workflow is not necessarily transparent nor consistent among studies. We share a workflow diagramming strategy that could provide coherency among practitioners. A false‐positive occupancy model is explored that accounts for misclassification errors and enables potential reduction in the number of confirmed detections. Simulations informed by real data were used to evaluate how much confirmation effort could be reduced without sacrificing site occupancy and detection error estimator bias and precision. We found even under a 50% reduction in total confirmation effort, estimator properties were reasonable for our assumed survey design, species‐specific parameter values, and desired precision. For transferability, a fully documented r package, OCacoustic, for implementing a false‐positive occupancy model is provided. Practitioners can apply OCacoustic to optimize their own study design (required sample sizes, number of visits, and confirmation scenarios) for properly implementing a false‐positive occupancy model with bat or other wildlife acoustic data. Additionally, our work highlights the importance of clearly defining research objectives and data processing strategies at the outset to align the study design with desired statistical inferences.     

Using occupancy modelling to compare environmental DNA to traditional field methods for regional‐scale monitoring of an endangered aquatic species

Environmental DNA (eDNA) monitoring approaches promise to greatly improve detection of rare, endangered and invasive species in comparison with traditional field approaches. Herein, eDNA approaches and traditional seining methods were applied at 29 research locations to compare method‐specific estimates of detection and occupancy probabilities for endangered tidewater goby (Eucyclogobius newberryi). At each location, multiple paired seine hauls and water samples for eDNA analysis were taken, ranging from two to 23 samples per site, depending upon habitat size. Analysis using a multimethod occupancy modelling framework indicated that the probability of detection using eDNA was nearly double (0.74) the rate of detection for seining (0.39). The higher detection rates afforded by eDNA allowed determination of tidewater goby occupancy at two locations where they have not been previously detected and at one location considered to be locally extirpated. Additionally, eDNA concentration was positively related to tidewater goby catch per unit effort, suggesting eDNA could potentially be used as a proxy for local tidewater goby abundance. Compared to traditional field sampling, eDNA provided improved occupancy parameter estimates and can be applied to increase management efficiency across a broad spatial range and within a diversity of habitats.     

Reliable eDNA detection and quantification of the European weather loach (Misgurnus fossilis)

The European weather loach (Misgurnus fossilis) is a cryptic and poorly known fish species of high conservation concern. The species is experiencing dramatic population collapses across its native range to the point of regional extinction. Although environmental DNA (eDNA)-based approaches offer clear advantages over conventional field methods for monitoring rare and endangered species, accurate detection and quantification remain difficult and quality assessment is often poorly incorporated. In this study, we developed and validated a novel digital droplet PCR (ddPCR) eDNA-based method for reliable detection and quantification, which allows accurate monitoring of M. fossilis across a number of habitat types. A dilution experiment under laboratory conditions allowed the definition of the limit of detection (LOD) and the limit of quantification (LOQ), which were set at concentrations of 0.07 and 0.14 copies μl^–1, respectively. A series of aquarium experiments revealed a significant and positive relationship between the number of individuals and the eDNA concentration measured. During a 3 year survey (2017–2019), we assessed 96 locations for the presence of M. fossilis in Flanders (Belgium). eDNA analyses on these samples highlighted 45% positive detections of the species. On the basis of the eDNA concentration per litre of water, only 12 sites appeared to harbour relatively dense populations. The other 31 sites gave a relatively weak positive signal that was typically situated below the LOQ. Combining sample-specific estimates of effective DNA quantity (Q_e) and conventional field sampling, we concluded that each of these weak positive sites still likely harboured the species and therefore they do not represent false positives. Further, only seven of the classified negative samples warrant additional sampling as our analyses identified a substantial risk of false-negative detections (i.e., type II errors) at these locations. Finally, we illustrated that ddPCR outcompetes conventional qPCR analyses, especially when target DNA concentrations are critically low, which could be attributed to a reduced sensitivity of ddPCR to inhibition effects, higher sample concentrations being accommodated and higher sensitivity obtained.     

On minimizing assignment errors and the trade‐off between false positives and negatives in parentage analysis

Genetic parentage analyses provide a practical means with which to identify parent–offspring relationships in the wild. In Harrison et al.'s study (2013a), we compare three methods of parentage analysis and showed that the number and diversity of microsatellite loci were the most important factors defining the accuracy of assignments. Our simulations revealed that an exclusion‐Bayes theorem method was more susceptible to false‐positive and false‐negative assignments than other methods tested. Here, we analyse and discuss the trade‐off between type I and type II errors in parentage analyses. We show that controlling for false‐positive assignments, without reporting type II errors, can be misleading. Our findings illustrate the need to estimate and report both the rate of false‐positive and false‐negative assignments in parentage analyses.     

Does accounting for imperfect detection improve species distribution models?

Models of species distributions are increasingly being used to address a variety of problems in conservation biology. In many applications, perfect or constant detectability of species, given presence, is assumed. While this problem has been acknowledged and addressed through the development of occupancy models, we still know little regarding whether addressing the potential for imperfect detection improves the predictive performance of species distribution models in nature. Here, we contrast logistic regression models of species occurrence that do not correct for detectability to hierarchical occupancy models that explicitly estimate and adjust for detectability, and maximum entropy models that attempt to circumvent the detectability problem by using data from known presence locations only. We use a large‐scale, long‐term monitoring database across western Montana and northern Idaho to contrast these models for nine landbird species that cover a broad spectrum in detectability. Overall, occupancy models were similar to or better than other approaches in terms of predictive accuracy, as measured by the Area Under the ROC Curve (AUC) and Kappa, with maximum entropy tending to provide the lowest predictive accuracy. Models varied in the types of errors associated with predictions, such that some model approaches may be preferred over others in certain situations. As expected, predictive performance varied across a gradient in species detectability, with logistic regression providing lower relative performance for less detectable species and Maxent providing lower performance for highly detectable species. We conclude by discussing the advantages and limitations to each approach for developing large‐scale species distribution models.     

Detection limits of quantitative and digital PCR assays and their influence in presence–absence surveys of environmental DNA

A set of universal guidelines is needed to determine the limit of detection (LOD) in PCR‐based analyses of low‐concentration DNA. In particular, environmental DNA (eDNA) studies require sensitive and reliable methods to detect rare and cryptic species through shed genetic material in environmental samples. Current strategies for assessing detection limits of eDNA are either too stringent or subjective, possibly resulting in biased estimates of species’ presence. Here, a conservative LOD analysis grounded in analytical chemistry is proposed to correct for overestimated DNA concentrations predominantly caused by the concentration plateau, a nonlinear relationship between expected and measured DNA concentrations. We have used statistical criteria to establish formal mathematical models for both quantitative and droplet digital PCR. To assess the method, a new Grass Carp (Ctenopharyngodon idella) TaqMan assay was developed and tested on both PCR platforms using eDNA in water samples. The LOD adjustment reduced Grass Carp occupancy and detection estimates while increasing uncertainty—indicating that caution needs to be applied to eDNA data without LOD correction. Compared to quantitative PCR, digital PCR had higher occurrence estimates due to increased sensitivity and dilution of inhibitors at low concentrations. Without accurate LOD correction, species occurrence and detection probabilities based on eDNA estimates are prone to a source of bias that cannot be reduced by an increase in sample size or PCR replicates. Other applications also could benefit from a standardized LOD such as GMO food analysis and forensic and clinical diagnostics.     

Quantification of mesocosm fish and amphibian species diversity via environmental DNA metabarcoding

Freshwater fauna are particularly sensitive to environmental change and disturbance. Management agencies frequently use fish and amphibian biodiversity as indicators of ecosystem health and a way to prioritize and assess management strategies. Traditional aquatic bioassessment that relies on capture of organisms via nets, traps and electrofishing gear typically has low detection probabilities for rare species and can injure individuals of protected species. Our objective was to determine whether environmental DNA (eDNA) sampling and metabarcoding analysis can be used to accurately measure species diversity in aquatic assemblages with differing structures. We manipulated the density and relative abundance of eight fish and one amphibian species in replicated 206‐L mesocosms. Environmental DNA was filtered from water samples, and six mitochondrial gene fragments were Illumina‐sequenced to measure species diversity in each mesocosm. Metabarcoding detected all nine species in all treatment replicates. Additionally, we found a modest, but positive relationship between species abundance and sequencing read abundance. Our results illustrate the potential for eDNA sampling and metabarcoding approaches to improve quantification of aquatic species diversity in natural environments and point the way towards using eDNA metabarcoding as an index of macrofaunal species abundance.     

Determining Occurrence Dynamics when False Positives Occur: Estimating the Range Dynamics of Wolves from Public Survey Data

Large-scale presence-absence monitoring programs have great promise for many conservation applications. Their value can be limited by potential incorrect inferences owing to observational errors, especially when data are collected by the public. To combat this, previous analytical methods have focused on addressing non-detection from public survey data. Misclassification errors have received less attention but are also likely to be a common component of public surveys, as well as many other data types. We derive estimators for dynamic occupancy parameters (extinction and colonization), focusing on the case where certainty can be assumed for a subset of detections. We demonstrate how to simultaneously account for non-detection (false negatives) and misclassification (false positives) when estimating occurrence parameters for gray wolves in northern Montana from 2007–2010. Our primary data source for the analysis was observations by deer and elk hunters, reported as part of the state’s annual hunter survey. This data was supplemented with data from known locations of radio-collared wolves. We found that occupancy was relatively stable during the years of the study and wolves were largely restricted to the highest quality habitats in the study area. Transitions in the occupancy status of sites were rare, as occupied sites almost always remained occupied and unoccupied sites remained unoccupied. Failing to account for false positives led to over estimation of both the area inhabited by wolves and the frequency of turnover. The ability to properly account for both false negatives and false positives is an important step to improve inferences for conservation from large-scale public surveys. The approach we propose will improve our understanding of the status of wolf populations and is relevant to many other data types where false positives are a component of observations.     

Estimating hazard ratios in cohort data with missing disease information due to death

In clinical and epidemiological studies information on the primary outcome of interest, that is, the disease status, is usually collected at a limited number of follow‐up visits. The disease status can often only be retrieved retrospectively in individuals who are alive at follow‐up, but will be missing for those who died before. Right‐censoring the death cases at the last visit (ad‐hoc analysis) yields biased hazard ratio estimates of a potential risk factor, and the bias can be substantial and occur in either direction. In this work, we investigate three different approaches that use the same likelihood contributions derived from an illness‐death multistate model in order to more adequately estimate the hazard ratio by including the death cases into the analysis: a parametric approach, a penalized likelihood approach, and an imputation‐based approach. We investigate to which extent these approaches allow for an unbiased regression analysis by evaluating their performance in simulation studies and on a real data example. In doing so, we use the full cohort with complete illness‐death data as reference and artificially induce missing information due to death by setting discrete follow‐up visits. Compared to an ad‐hoc analysis, all considered approaches provide less biased or even unbiased results, depending on the situation studied. In the real data example, the parametric approach is seen to be too restrictive, whereas the imputation‐based approach could almost reconstruct the original event history information.