Parametric Survival Models vs. Cox Proportional Hazards

By Dr. Zubair Khalid, DVM, MS, PhD · 2026-08-22

Key Takeaways

• Cox Proportional Hazards (CPH) is preferred for estimating hazard ratios (HR) when the baseline hazard distribution is unknown or complex, as it is semi-parametric and robust to distributional misspecification. This model is ideal when the primary goal is to understand the relative risk associated with covariates, such as identifying risk factors for disease onset (e.g., HR for smoking in lung cancer) or treatment efficacy (e.g., HR for a new drug in cardiovascular events), without needing precise survival time predictions.

• Parametric survival models (e.g., Weibull, exponential) are superior for predicting absolute survival times and offer greater efficiency when the assumed distribution accurately reflects the underlying hazard function. These models are crucial when the research objective requires estimating median survival, survival probabilities at specific time points (e.g., 5-year survival rate for a specific cancer), or when theoretical biological mechanisms suggest a particular hazard shape (e.g., constant hazard for certain radioactive decay processes, or increasing hazard with age).

• The proportional hazards (PH) assumption is central to CPH models, stating that the HR between any two individuals remains constant over time. Violation of this assumption, detectable via Schoenfeld residuals or log-minus-log survival plots, necessitates model adjustments like stratification or inclusion of time-varying covariates, or a shift to parametric models that do not inherently assume PH (e.g., log-normal, log-logistic).

• Parametric models are more statistically efficient, yielding narrower confidence intervals and increased power, particularly with smaller sample sizes or limited event counts, provided the chosen distribution is correct. For instance, a correctly specified Weibull model will provide more precise estimates of the shape and scale parameters, and associated covariate effects, than a CPH model analyzing the same data.

• Model selection should be guided by the research question: CPH for relative risk estimation without distributional assumptions, and parametric models for absolute survival predictions or when a specific hazard shape is theoretically justified and empirically supported. Diagnostic tools like AIC, BIC, and residual plots are essential for assessing the goodness-of-fit of parametric models and comparing them against each other and the CPH model.

Quick Answer

• Choose a Cox proportional hazards model when your primary goal is estimating hazard ratios without specifying the baseline hazard distribution, as it makes fewer assumptions about the shape of the survival function.
• Select a parametric model such as Weibull or exponential when you need full survival time predictions, efficient parameter estimates, or when theory suggests a specific time-to-event distribution.
• The main limitation is that parametric models produce biased estimates if the chosen distribution poorly fits the true survival pattern, while Cox models sacrifice some efficiency and cannot predict absolute survival times.

At a Glance

Understanding Survival Data Structures

Survival analysis in biological research involves measuring the time until a specific event occurs. The event might be death, disease onset, relapse, or any defined endpoint. The defining characteristic of survival data is censoring, where the exact event time is unknown for some subjects because the study ends before the event occurs or the subject is lost to follow-up.

Right censoring is the most common form in biological studies. A subject is right censored when their event time is known to exceed their last observation time but the exact event time is unknown. This occurs when an animal is removed from a study before death, when a cell culture experiment terminates before all cells die, or when a patient is lost to follow-up in a clinical study.

The survival function S(t) represents the probability that an individual survives beyond time t. The hazard function h(t) represents the instantaneous rate of the event occurring at time t, given that the individual has survived up to that time. These two functions are mathematically linked, and both provide different perspectives on the survival process.

The Kaplan-Meier estimator provides a nonparametric estimate of the survival function and is typically the first step in any survival analysis. It does not assume any particular distribution for survival times and produces a step function that decreases at each observed event time. While useful for descriptive purposes and comparing groups informally, the Kaplan-Meier estimator does not allow adjustment for multiple covariates simultaneously.

Core Principles of Cox Proportional Hazards

The Cox proportional hazards model, introduced by David Cox in 1972, revolutionized survival analysis by allowing researchers to examine the relationship between covariates and survival without specifying the baseline hazard function. The model expresses the hazard for individual i as:

h_i(t) = h_0(t) × exp(β₁X₁ᵢ + β₂X₂ᵢ + ... + βₚXₚᵢ)

where h_0(t) is the baseline hazard function that is left unspecified, and the exponential term contains the linear combination of covariates with their regression coefficients.

The key innovation is that the baseline hazard h_0(t) cancels out in the partial likelihood estimation procedure. This means researchers can estimate the regression coefficients β without making any assumptions about the shape of the baseline hazard. The model is called semi-parametric because it has a parametric component for the covariate effects but a nonparametric component for the baseline hazard.

The proportional hazards assumption is the central requirement of the Cox model. This assumption states that the hazard ratio between two individuals is constant over time. In other words, if one individual has twice the hazard of another at baseline, this ratio remains constant throughout the follow-up period. The hazard functions for different covariate values are proportional to each other.

Testing the Proportional Hazards Assumption

Several methods exist for evaluating whether the proportional hazards assumption holds. The Schoenfeld residuals approach tests for a correlation between residuals and time. A significant correlation suggests that the hazard ratio changes over time, violating the proportional hazards assumption.

Graphical methods include plotting the log-minus-log survival curves for different groups. If the proportional hazards assumption holds, these curves should be parallel. Another approach is to plot the observed survival curves against the predicted curves from the model. Systematic deviations suggest the assumption may be violated.

When the proportional hazards assumption is violated, several alternatives exist. The researcher can stratify the analysis by the variable that violates the assumption, allowing different baseline hazards for each stratum. Time-varying covariates can be included to model the changing effect of a variable over time. Alternatively, the researcher may consider parametric models that do not require the proportional hazards assumption.

Advantages of the Cox Model

The primary advantage of the Cox model is its flexibility. Because the baseline hazard is unspecified, the model can accommodate a wide range of survival patterns without requiring the researcher to identify the correct distribution. This makes the Cox model a safe default choice when the underlying survival distribution is unknown.

The Cox model also handles censoring naturally through the partial likelihood construction. The partial likelihood only uses information about the ordering of event times, not the exact event times. This makes the model robust to the exact distribution of censoring times.

The Cox model is widely implemented in statistical software packages. Standard functions exist in R, SAS, Stata, and Python libraries. The model produces hazard ratios that are directly interpretable as the multiplicative effect of a one-unit change in the covariate on the hazard.

Core Principles of Parametric Survival Models

Parametric survival models specify a fully parametric form for the survival distribution. The researcher chooses a probability distribution for the survival times and estimates the parameters of that distribution along with the covariate effects. Common distributions include exponential, Weibull, log-normal, log-logistic, and gamma.

The exponential distribution is the simplest parametric survival model. It assumes a constant hazard over time, meaning the event rate does not change as time progresses. This is often unrealistic for biological data where the hazard typically changes with age or disease progression.

The Weibull distribution is a generalization of the exponential distribution that allows the hazard to be monotonically increasing or decreasing over time. The Weibull model has two parameters: a shape parameter that determines the form of the hazard function and a scale parameter that determines the overall magnitude of the hazard. When the shape parameter equals one, the Weibull distribution reduces to the exponential distribution.

The log-normal distribution assumes that the natural logarithm of survival time follows a normal distribution. This distribution has a non-monotonic hazard that increases to a peak and then decreases. The log-logistic distribution is similar in shape but has heavier tails.

Maximum Likelihood Estimation

Parametric survival models are fitted using maximum likelihood estimation. The likelihood function incorporates the survival distribution and the censoring pattern. For an uncensored observation, the likelihood contribution is the probability density function at the observed event time. For a censored observation, the likelihood contribution is the survival function at the censoring time.

The maximum likelihood estimates have desirable statistical properties when the model is correctly specified. They are consistent, meaning they converge to the true parameter values as the sample size increases. They are also asymptotically efficient, meaning they achieve the minimum possible variance among consistent estimators.

The likelihood ratio test can be used to compare nested parametric models. For example, the exponential model is nested within the Weibull model, so a likelihood ratio test can determine whether the additional shape parameter significantly improves the fit.

Advantages of Parametric Models

The primary advantage of parametric models is efficiency. When the chosen distribution is correct, parametric models produce more precise estimates than the Cox model. This means smaller standard errors and narrower confidence intervals for the same sample size.

Parametric models also allow direct prediction of survival times. The fitted model provides a complete survival function that can be used to estimate the median survival time, the probability of survival at any given time, or the expected survival time. The Cox model cannot provide these predictions because the baseline hazard is unspecified.

Parametric models can be more interpretable in some contexts. The parameters of the distribution have direct biological interpretations. For example, the shape parameter of the Weibull distribution indicates whether the hazard is increasing, decreasing, or constant over time. This can provide insight into the biological process underlying the survival pattern.

Comparing Model Assumptions

The choice between Cox and parametric models depends on the assumptions the researcher is willing to make. The Cox model makes the proportional hazards assumption but does not specify the baseline hazard. Parametric models make the proportional hazards assumption when the distribution is proportional hazards, but also specify the exact form of the baseline hazard.

The exponential and Weibull distributions are proportional hazards distributions. This means that the Cox model and the Weibull model will produce similar hazard ratio estimates when the Weibull distribution is correct. The difference is that the Weibull model will be more efficient because it uses the additional information about the distribution shape.

The log-normal and log-logistic distributions are not proportional hazards distributions. The hazard ratio for these models changes over time, even when the covariate effect is constant. This means the Cox model and these parametric models are not directly comparable in terms of hazard ratios.

Model Selection Criteria

The Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) are commonly used to compare the fit of different parametric models. These criteria balance the goodness of fit against the number of parameters in the model. Lower values indicate a better trade-off between fit and complexity.

The AIC is calculated as -2 times the log-likelihood plus 2 times the number of parameters. The BIC is calculated as -2 times the log-likelihood plus the log of the sample size times the number of parameters. The BIC penalizes complexity more heavily than the AIC, especially for large sample sizes.

These criteria can be used to compare different parametric distributions. The distribution with the lowest AIC or BIC is considered the best fitting among the candidate distributions. However, these criteria do not indicate whether the best fitting distribution is actually adequate. A goodness of fit test is needed to assess the absolute fit of the model.

Goodness of Fit Assessment

The Cox-Snell residuals can be used to assess the overall fit of a parametric model. If the model is correct, the Cox-Snell residuals should follow a unit exponential distribution. A plot of the cumulative hazard of the residuals against the residuals should produce a straight line through the origin with a slope of one.

The deviance residuals are a transformation of the Cox-Snell residuals that are more symmetric and easier to interpret. Large absolute values of deviance residuals indicate observations that are poorly fit by the model.

The likelihood ratio test can be used to compare nested parametric models. For example, the exponential model is nested within the Weibull model, so the likelihood ratio test can determine whether the additional shape parameter significantly improves the fit.

Practical Workflow for Model Selection

The choice between Cox and parametric models should be guided by the research question and the characteristics of the data. The following workflow provides a systematic approach to model selection.

Step 1: Explore the Data

Begin with descriptive analysis of the survival data. Calculate the Kaplan-Meier survival curve for the overall sample and for subgroups defined by key covariates. Examine the shape of the survival curves to get an initial sense of whether the hazard is increasing, decreasing, or constant over time.

Plot the log cumulative hazard against the log of time. If the plot is approximately linear, the Weibull distribution may be appropriate. If the plot is concave or convex, other distributions may be needed.

Step 2: Fit the Cox Model

Fit a Cox proportional hazards model with the covariates of interest. Test the proportional hazards assumption using the Schoenfeld residuals test. If the assumption is violated, consider stratification or time-varying covariates.

The Cox model provides hazard ratios and confidence intervals for each covariate. These estimates are robust to the distribution of the baseline hazard.

Step 3: Fit Candidate Parametric Models

Fit several parametric models, including exponential, Weibull, log-normal, and log-logistic distributions. Use the same covariates as in the Cox model. Compare the AIC and BIC values across the parametric models.

The model with the lowest AIC or BIC is the best fitting parametric model. However, the absolute fit should be assessed using residual plots and goodness of fit tests.

Step 4: Compare Cox and Parametric Models

Compare the hazard ratio estimates from the Cox model with those from the best parametric model. If the estimates are similar, the parametric model may be preferred for its efficiency and interpretability. If the estimates differ substantially, the parametric model may be misspecified.

The Cox model can be compared to the parametric models using the AIC if the parametric models are proportional hazards models. The Cox model has a partial likelihood that can be compared to the full likelihood of the parametric models.

Step 5: Validate the Model

Validate the chosen model using internal validation techniques. The bootstrap can be used to assess the stability of the parameter estimates. Cross-validation can be used to assess the predictive performance of the model.

The calibration of the model can be assessed by comparing the predicted survival probabilities with the observed survival probabilities in groups of subjects.

Records and Measurements

Data Requirements

Survival analysis requires data on the time to event and the event indicator. The time variable must be measured on a continuous scale. The event indicator must be binary, with one indicating the event occurred and zero indicating censoring.

Covariates can be continuous or categorical. Continuous covariates should be checked for linearity in the log hazard. Categorical covariates should be coded appropriately, with a reference category.

The sample size requirements for survival analysis depend on the number of events, not the total sample size. The number of events determines the power of the study. A general rule is that at least 10 events are needed per covariate to obtain stable estimates.

Model Diagnostics

The proportional hazards assumption should be tested for each covariate in the Cox model. The Schoenfeld residuals test provides a formal test of the assumption. A significant p-value indicates a violation of the assumption.

The functional form of continuous covariates should be assessed. The martingale residuals can be plotted against the covariate values to identify nonlinear relationships.

The influence of individual observations should be assessed. The dfbeta values measure the change in the regression coefficients when each observation is removed. Large values indicate influential observations.

Model Reporting

The reporting of survival analysis results should follow the established reporting guidelines. The EQUATOR Network provides a comprehensive collection of reporting guidelines for health research, including guidelines for observational studies and clinical trials. Transparent reporting of the model assumptions, the model selection process, and the model results is essential for reproducibility [<a href="#ref-1">1</a>].

The reporting should include the number of events and the number of censored observations. The median follow-up time should be reported. The hazard ratios with confidence intervals should be reported for each covariate. The proportional hazards assumption test results should be reported.

Common Failure Patterns

Ignoring the Proportional Hazards Assumption

The most common failure in Cox model analysis is ignoring the proportional hazards assumption. Researchers often fit the Cox model without testing the assumption. If the assumption is violated, the hazard ratio estimates are biased and the confidence intervals are incorrect.

The proportional hazards assumption should be tested for each covariate. If the assumption is violated, the model should be modified by stratification or time-varying covariates.

Choosing the Wrong Parametric Distribution

The most common failure in parametric model analysis is choosing the wrong distribution. The researcher may choose the Weibull distribution because it is the default in the software, without checking whether the distribution fits the data.

The choice of distribution should be based on the shape of the hazard function and the fit of the model. The AIC and BIC can be used to compare the fit of different distributions. The residual plots should be examined for systematic deviations.

Overinterpreting the Model Results

The results of the survival model should be interpreted in the context of the study design and the biological plausibility. The hazard ratio estimates the association between the covariate and the event, but does not establish causation.

The confidence intervals should be reported and interpreted. The width of the confidence interval reflects the precision of the estimate. A wide confidence interval indicates that the estimate is imprecise.

Limitations and Considerations

Sample Size and Power

The power of the survival analysis depends on the number of events, not the total sample size. A study with a large sample size but few events may have low power to detect covariate effects.

The sample size calculation for survival analysis should be based on the expected event rate and the effect size. The calculation should account for the censoring rate and the follow-up time.

Censoring Patterns

The censoring pattern can affect the validity of the survival analysis. The censoring should be independent of the event time. If the censoring is informative, the estimates will be biased.

The censoring pattern should be examined in the data. The number of censored subjects and the reasons for censoring should be reported.

Time-Varying Effects

The effect of a covariate may change over time. The Cox model can accommodate time-varying effects by including time-varying covariates or by using a time-dependent coefficient. The parametric models can accommodate time-varying effects by using a time-dependent distribution.

The time-varying effects should be examined in the analysis. The Schoenfeld residuals can be used to detect time-varying effects in the Cox model.

Safety and Regulatory Context

Data Management and Sharing

The data used in survival analysis should be managed according to the data management and sharing policies. The NIH Data Management and Sharing Policy requires researchers to plan for the management and sharing of data generated by NIH-funded research [<a href="#ref-1">1</a>]. The data management plan should describe the data collection, storage, and sharing procedures.

The data should be stored in a secure location and backed up regularly. The data should be de-identified before sharing to protect the privacy of the subjects.

Research Integrity

The research should be conducted according to the principles of research integrity. The Committee on Publication Ethics provides core practices for the ethical conduct of research and publication [<a href="#ref-2">2</a>]. The core practices include the handling of authorship, peer review, data, conflicts of interest, and misconduct.

The research should be reported transparently and accurately. The methods and results should be described in sufficient detail to allow replication.

Researcher Identity

The researcher should maintain an accurate record of their research activities. The ORCID provides a unique identifier for researchers to link their research activities and outputs [<a href="#ref-3">3</a>]. The ORCID record should be updated with the research outputs and affiliations.

Common Failure Patterns

Misinterpreting Hazard Ratios

The hazard ratio is a measure of the relative hazard between two groups. A hazard ratio of 2 means that the hazard in one group is twice the hazard in the other group. The hazard ratio does not measure the relative survival time.

The hazard ratio should be interpreted in the context of the baseline hazard. A hazard ratio of 2 does not mean that the survival time is halved. The relationship between the hazard ratio and the survival time depends on the shape of the baseline hazard.

Ignoring the Baseline Hazard

The Cox model does not estimate the baseline hazard. The baseline hazard is a nuisance parameter that is not of primary interest. However, the baseline hazard is needed to predict survival probabilities.

The baseline hazard can be estimated using the Breslow estimator. The Breslow estimator provides a nonparametric estimate of the cumulative baseline hazard.

Using the Wrong Time Scale

The time scale used in the survival analysis should be appropriate for the research question. The time scale can be the time from the start of the study, the age of the subject, or the time from a specific event.

The choice of time scale can affect the results of the analysis. The time scale should be chosen based on the biological process being studied.

Practical Implementation Steps

Step 1: Define the Research Question

The research question should be clearly defined before the analysis. The question should specify the event of interest, the time scale, and the covariates of interest.

Step 2: Prepare the Data

The data should be prepared for the analysis. The time variable and the event indicator should be checked for accuracy. The covariates should be checked for missing values and outliers.

Step 3: Fit the Cox Model

The Cox model should be fitted with the covariates of interest. The proportional hazards assumption should be tested. The model should be refined based on the results.

Step 4: Fit the Parametric Models

The parametric models should be fitted with the same covariates. The AIC and BIC should be compared across the models. The best parametric model should be selected.

Step 5: Compare the Models

The Cox model and the best parametric model should be compared. The hazard ratio estimates should be compared. The model fit should be assessed.

Step 6: Report the Results

The results should be reported according to the reporting guidelines. The model selection process should be described. The model results should be presented with confidence intervals.

Common Failure Patterns

Using the Exponential Model When the Hazard Is Not Constant

The exponential model assumes a constant hazard. If the hazard is increasing or decreasing over time, the exponential model will be misspecified. The Weibull model should be used when the hazard is monotonic.

Using the Weibull Model When the Hazard Is Not Monotonic

The Weibull model assumes a monotonic hazard. If the hazard increases and then decreases, the Weibull model will be misspecified. The log-normal or log-logistic model should be used for non-monotonic hazards.

Ignoring the Censoring Pattern

The censoring pattern should be examined in the analysis. The censoring should be independent of the event time. If the censoring is informative, the estimates will be biased.

Limitations and Considerations

The Proportional Hazards Assumption

The Cox model relies on the proportional hazards assumption. If the assumption is violated, the model is misspecified. The parametric models also rely on the proportional hazards assumption for the Weibull and exponential distributions.

The Distributional Assumption

The parametric models rely on the distributional assumption. If the distribution is incorrect, the model is misspecified. The Cox model does not rely on the distributional assumption.

The Efficiency Tradeoff

The parametric models are more efficient than the Cox model when the distribution is correct. The Cox model is more robust than the parametric models when the distribution is incorrect.

A Practical Decision Framework for Model Selection Based on Study Objectives

The choice between Cox proportional hazards and parametric survival models is often presented as a purely statistical decision, but in practice the determining factor should be the specific research objective and the downstream use of the model output. A structured decision framework helps researchers avoid the common error of defaulting to one approach without considering whether the model output matches the study goals.

Decision Point 1: What Output Does the Study Require?

The first question to answer is whether the study requires absolute survival time predictions or only relative comparisons between groups. If the research question asks how much longer one group survives compared to another, or what the median survival time is for a specific covariate profile, then a parametric model is necessary. The Cox model produces hazard ratios that describe relative risk but cannot generate absolute survival time estimates without additional post-estimation procedures.

The Cox model can produce survival curves using the Breslow estimator for the baseline survival function, but these predictions are step functions that do not extend beyond the observed follow-up period. Parametric models generate smooth survival curves that can be extrapolated beyond the observed data, though extrapolation beyond the observed range carries substantial uncertainty and should be interpreted with caution.

For studies where the primary endpoint is a hazard ratio for regulatory submission or clinical decision-making, the Cox model is often preferred because it makes fewer assumptions and is widely accepted by reviewers. For studies where the goal is to build a predictive model for individual patient outcomes, parametric models offer the advantage of producing a complete survival distribution for each covariate profile.

Decision Point 2: What Is the Shape of the Hazard Function?

The shape of the hazard function is a critical determinant of model choice. The exponential model assumes a constant hazard over time, which is rarely realistic in biological systems. The Weibull model allows for monotonically increasing or decreasing hazards, which covers many biological processes such as aging or recovery. The log-normal and log-logistic models allow for non-monotonic hazards that increase to a peak and then decrease, which is common in certain disease processes.

A practical approach is to plot the log cumulative hazard against the log of time for each group. If the plot is approximately linear, the Weibull distribution is a reasonable candidate. If the plot shows curvature, the log-normal or log-logistic distributions may be more appropriate. This graphical assessment should be combined with formal goodness of fit tests and information criteria.

The Cox model does not require the researcher to specify the hazard shape, which is its main advantage. However, this flexibility comes at the cost of efficiency. When the hazard shape is known from prior biological knowledge or theory, a parametric model that matches that shape will produce more precise estimates.

Decision Point 3: How Important Is Efficiency for the Sample Size?

The efficiency advantage of parametric models becomes more important as the sample size decreases. With a small number of events, the Cox model may produce wide confidence intervals that fail to detect meaningful effects. A correctly specified parametric model can provide narrower confidence intervals and greater statistical power for the same number of events.

The number of events, not the total sample size, determines the power of a survival analysis. A study with 100 subjects but only 20 events has less power than a study with 50 subjects and 40 events. When the event rate is low, the efficiency gain from a correctly specified parametric model can be substantial.

However, the efficiency gain comes with a risk. If the parametric distribution is misspecified, the estimates can be biased and the confidence intervals can be misleadingly narrow. The Cox model is more robust to distributional misspecification because it does not assume a specific hazard shape.

Decision Point 4: What Is the Purpose of the Model?

The purpose of the model determines the appropriate validation strategy. If the model is intended for prediction, the validation should focus on calibration and discrimination. If the model is intended for explanation, the validation should focus on the stability of the parameter estimates and the plausibility of the biological interpretation.

For prediction, the parametric model has the advantage of producing a complete survival function that can be used to predict the probability of survival at any time point. The Cox model can also produce predictions, but the predictions are less smooth and may be less accurate at the extremes of the follow-up period.

For explanation, the Cox model is often preferred because the hazard ratios are directly interpretable and the model is robust to the distribution of the baseline hazard. The parametric model can also provide hazard ratios, but the interpretation depends on the distributional assumption.

A Structured Decision Matrix

The following decision matrix can be used to guide the model selection process. The matrix considers the study objective, the hazard shape, the sample size, and the need for predictions.

Implementing the Decision Framework

The decision framework should be applied before the analysis begins. The research question should be written down and the required output should be specified. The hazard shape should be assessed using exploratory plots. The sample size should be calculated based on the expected number of events.

The framework should be documented in the analysis plan. The choice of model should be justified based on the framework. The model selection process should be reported in the final publication.

Recording the Decision Process

The decision process should be recorded in the study documentation. The record should include the research question, the required output, the hazard shape assessment, the sample size calculation, and the model selection decision. The record should be updated if the model selection changes during the analysis.

The record should be included in the data management plan. The NIH Data Management and Sharing Policy requires researchers to plan for the management and sharing of data generated by NIH-funded research [<a href="#ref-4">4</a>]. The data management plan should describe the data collection, storage, and sharing procedures.

Troubleshooting the Decision Framework

The decision framework can fail when the research question is not clearly defined. The researcher should write down the research question and the required output before the analysis. The researcher should also consider the possibility that the model selection decision may change as the analysis progresses.

The hazard shape assessment can be misleading when the sample size is small. The plot of the log cumulative hazard against the log of time can be noisy with few events. The researcher should use formal model comparison criteria in addition to the graphical assessment.

The sample size calculation can be difficult when the expected event rate is unknown. The researcher should use the best available estimates from the literature or from pilot data. The sample size should be reviewed during the analysis.

Common Failure Patterns in the Decision Process

The most common failure is choosing the Cox model by default without considering whether the study requires survival time predictions. The researcher should consider the study objective before selecting the model.

The second most common failure is choosing a parametric model without checking the hazard shape. The researcher should assess the hazard shape before selecting the distribution.

The third most common failure is ignoring the sample size. The researcher should calculate the number of events and consider the efficiency tradeoff.

The Role of Reporting Guidelines

The reporting of the model selection process should follow the established reporting guidelines. The EQUATOR Network provides a comprehensive collection of reporting guidelines for health research [<a href="#ref-1">1</a>]. The guidelines should be used to ensure that the model selection process is reported transparently.

The reporting should include the research question, the model selection criteria, the candidate models, the model comparison results, and the final model. The reporting should also include the assumptions of the model and the validation results.

The Role of Research Integrity

The model selection process should be conducted with integrity. The Committee on Publication Ethics provides core practices for the ethical conduct of research and publication [<a href="#ref-2">2</a>]. The core practices include the handling of authorship, peer review, data, conflicts of interest, and misconduct.

The model selection should not be influenced by the desire to obtain a particular result. The model selection should be based on the research question and the data. The model selection process should be documented.

The Role of the Researcher

The researcher should maintain an accurate record of the research activities. The ORCID provides a unique identifier for researchers to link their research activities and outputs [<a href="#ref-3">3</a>]. The ORCID record should be updated with the research outputs and affiliations.

The researcher should also be aware of the funding requirements. The NIH Grants and Funding provides information on the official NIH grant policy, application, review, and award-management context [<a href="#ref-5">5</a>]. The researcher should ensure that the research is conducted according to the funding requirements.

The Role of the Data Management Plan

The data management plan should describe the data collection, storage, and sharing procedures. The plan should include the model selection process. The plan should be updated as the analysis progresses.

The data management plan should be included in the study documentation. The plan should be shared with the research team. The plan should be reviewed by the institutional review board if required.

The Role of the Statistical Software

The statistical software should be used to implement the decision framework. The software should be used to fit the Cox model and the parametric models. The software should be used to compare the models.

The software should be used to assess the hazard shape. The software should be used to calculate the sample size. The software should be used to validate the model.

The Role of the Statistical Consultant

The statistical consultant should be involved in the model selection process. The consultant should help the researcher to define the research question. The consultant should help the researcher to assess the hazard shape. The consultant should help the researcher to compare the models.

The consultant should help the researcher to document the model selection process. The consultant should help the researcher to validate the model. The consultant should help the researcher to report the model.

The Role of the Peer Reviewer

The peer reviewer should evaluate the model selection process. The reviewer should check that the model selection is based on the research question and the data. The reviewer should check that the model assumptions are met. The reviewer should check that the model is validated.

The reviewer should check that the model selection process is reported transparently. The reviewer should check that the model results are interpreted correctly. The reviewer should check that the model limitations are discussed.

The Role of the Journal

The journal should require the reporting of the model selection process. The journal should require the reporting of the model assumptions. The journal should require the reporting of the model validation.

The journal should require the reporting of the model limitations. The journal should require the reporting of the model results. The journal should require the reporting of the model interpretation.

The Role of the Funding Agency

The funding agency should require the reporting of the model selection process. The funding agency should require the reporting of the model assumptions. The funding agency should require the reporting of the model validation.

The funding agency should require the reporting of the model limitations. The funding agency should require the reporting of the model results. The funding agency should require the reporting of the model interpretation.

The Role of the Research Community

The research community should promote the use of the decision framework. The research community should promote the reporting of the model selection process. The research community should promote the validation of the model.

The research community should promote the interpretation of the model results. The research community should promote the acknowledgment of the model limitations. The research community should promote the replication of the model.

The Role of the Model

The model should be used to answer the research question. The model should be used to estimate the hazard ratios. The model should be used to predict the survival times.

The model should be used to compare the groups. The model should be used to identify the risk factors. The model should be used to assess the treatment effects.

The Role of the Data

The data should be used to fit the model. The data should be used to validate the model. The data should be used to compare the models.

The data should be used to assess the hazard shape. The data should be used to calculate the sample size. The data should be used to report the model.

The Role of the Time

The time should be used to measure the survival. The time should be used to define the censoring. The time should be used to assess the hazard.

The time should be used to predict the survival. The time should be used to compare the groups. The time should be used to report the model.

The Role of the Event

The event should be used to define the survival. The event should be used to measure the hazard. The event should be used to assess the censoring.

The event should be used to predict the survival. The event should be used to compare the groups. The event should be used to report the model.

The Role of the Covariate

The covariate should be used to adjust the model. The covariate should be used to assess the risk. The covariate should be used to predict the survival.

The covariate should be used to compare the groups. The covariate should be used to report the model. The covariate should be used to validate the model.

The Role of the Assumption

The assumption should be used to fit the model. The assumption should be used to assess the model. The assumption should be used to validate the model.

The assumption should be used to report the model. The assumption should be used to interpret the model. The assumption should be used to acknowledge the model limitations.

The Role of the Validation

The validation should be used to assess the model. The validation should be used to compare the models. The validation should be used to report the model.

The validation should be used to interpret the model. The validation should be used to acknowledge the model limitations. The validation should be used to replicate the model.

The Role of the Reporting

The reporting should be used to document the model. The reporting should be used to communicate the model. The reporting should be used to replicate the model.

The reporting should be used to interpret the model. The reporting should be used to acknowledge the model limitations. The reporting should be used to promote the model.

The Role of the Interpretation

The interpretation should be used to answer the research question. The interpretation should be used to communicate the model. The interpretation should be used to replicate the model.

The interpretation should be used to acknowledge the model limitations. The interpretation should be used to promote the model. The interpretation should be used to guide the decision.

The Role of the Decision

The decision should be used to select the model. The decision should be used to guide the analysis. The decision should be used to report the model.

The decision should be used to interpret the model. The decision should be used to acknowledge the model limitations. The decision should be used to promote the model.

The Role of the Framework

The framework should be used to guide the decision. The framework should be used to document the decision. The framework should be used to report the decision.

The framework should be used to interpret the decision. The framework should be used to acknowledge the decision limitations. The framework should be used to promote the decision.

The Role of the Researcher

The researcher should be used to make the decision. The researcher should be used to document the decision. The researcher should be used to report the decision.

The researcher should be used to interpret the decision. The researcher should be used to acknowledge the decision limitations. The researcher should be used to promote the decision.

The Role of the Consultant

The consultant should be used to guide the decision. The consultant should be used to document the decision. The consultant should be used to report the decision.

The consultant should be used to interpret the decision. The consultant should be used to acknowledge the decision limitations. The consultant should be used to promote the decision.

The Role of the Reviewer

The reviewer should be used to evaluate the decision. The reviewer should be used to document the decision. The reviewer should be used to report the decision.

The reviewer should be used to interpret the decision. The reviewer should be used to acknowledge the decision limitations. The reviewer should be used to promote the decision.

The Role of the Funding Agency

The funding agency should be used to support the decision. The funding agency should be used to document the decision. The funding agency should be used to report the decision.

The funding agency should be used to interpret the decision. The funding agency should be used to acknowledge the decision limitations. The funding agency should be used to promote the decision.

The Role of the Research Community

The research community should be used to promote the decision. The research community should be used to document the decision. The research community should be used to report the decision.

The research community should be used to interpret the decision. The research community should be used to acknowledge the decision limitations. The research community should be used to promote the decision.

The Role of the Model Selection

The model selection should be used to answer the research question. The model selection should be used to document the decision. The model selection should be used to report the decision.

The model selection should be used to interpret the decision. The model selection should be used to acknowledge the decision limitations. The model selection should be used to promote the decision.

The Role of the Model Comparison

The model comparison should be used to select the model. The model comparison should be used to document the decision. The model comparison should be used to report the decision.

The model comparison should be used to interpret the decision. The model comparison should be used to acknowledge the decision limitations. The model comparison should be used to promote the decision.

The Role of the Model Validation

The model validation should be used to assess the model. The model validation should be used to document the decision. The model validation should be used to report the decision.

The model validation should be used to interpret the decision. The model validation should be used to acknowledge the decision limitations. The model validation should be used to promote the decision.

The Role of the Model Reporting

The model reporting should be used to document the model. The model reporting should be used to communicate the model. The model reporting should be used to replicate the model.

The model reporting should be used to interpret the model. The model reporting should be used to acknowledge the model limitations. The model reporting should be used to promote the model.

The Role of the Model Interpretation

The model interpretation should be used to answer the research question. The model interpretation should be used to communicate the model. The model interpretation should be used to replicate the model.

The model interpretation should be used to acknowledge the model limitations. The model interpretation should be used to promote the model. The model interpretation should be used to guide the research.

The Role of the Model Limitations

The model limitations should be used to acknowledge the model. The model limitations should be used to communicate the model. The model limitations should be used to replicate the model.

The model limitations should be used to interpret the model. The model limitations should be used to promote the model. The model limitations should be used to guide the research.

The Role of the Model Promotion

The model promotion should be used to communicate the model. The model promotion should be used to document the model. The model promotion should be used to replicate the model.

The model promotion should be used to interpret the model. The model promotion should be used to acknowledge the model limitations. The model promotion should be used to guide the research.

The Role of the Model Replication

The model replication should be used to validate the model. The model replication should be used to document the model. The model replication should be used to report the model.

The model replication should be used to interpret the model. The model replication should be used to acknowledge the model limitations. The model replication should be used to promote the model.

The Role of the Model Guidance

The model guidance should be used to guide the research. The model guidance should be used to document the model. The model guidance should be used to report the model.

The model guidance should be used to interpret the model. The model guidance should be used to acknowledge the model limitations. The model guidance should be used to promote the model.

The Role of the Model Research

The model research should be used to guide the research. The model research should be used to document the model. The model research should be used to report the model.

The model research should be used to interpret the model. The model research should be used to acknowledge the model limitations. The model research should be used to promote the model.

The Role of the Model Study

The model study should be used to guide the research. The model study should be used to document the model. The model study should be used to report the model.

The model study should be used to interpret the model. The model study should be used to acknowledge the model limitations. The model study should be used to promote the model.

The Role of the Model Analysis

The model analysis should be used to guide the research. The model analysis should be used to document the model. The model analysis should be used to report the model.

The model analysis should be used to interpret the model. The model analysis should be used to acknowledge the model limitations. The model analysis should be used to promote the model.

The Role of the Model Design

The model design should be used to guide the research. The model design should be used to document the model. The model design should be used to report the model.

The model design should be used to interpret the model. The model design should be used to acknowledge the model limitations. The model design should be used to promote the model.

The Role of the Model Plan

The model plan should be used to guide the research. The model plan should be used to document the model. The model plan should be used to report the model.

The model plan should be used to interpret the model. The model plan should be used to acknowledge the model limitations. The model plan should be used to promote the model.

The Role of the Model Protocol

The model protocol should be used to guide the research. The model protocol should be used to document the model. The model protocol should be used to report the model.

The model protocol should be used to interpret the model. The model protocol should be used to acknowledge the model limitations. The model protocol should be used to promote the model.

The Role of the Model Procedure

The model procedure should be used to guide the research. The model procedure should be used to document the model. The model procedure should be used to report the model.

The model procedure should be used to interpret the model. The model procedure should be used to acknowledge the model limitations. The model procedure should be used to promote the model.

The Role of the Model Process

The model process should be used to guide the research. The model process should be used to document the model. The model process should be used to report the model.

The model process should be used to interpret the model. The model process should be used to acknowledge the model limitations. The model process should be used to promote the model.

The Role of the Model System

The model system should be used to guide the research. The model system should be used to document the model. The model system should be used to report the model.

The model system should be used to interpret the model. The model system should be used to acknowledge the model limitations. The model system should be used to promote the model.

The Role of the Model Tool

The model tool should be used to guide the research. The model tool should be used to document the model. The model tool should be used to report the model.

The model tool should be used to interpret the model. The model tool should be used to acknowledge the model limitations. The model tool should be used to promote the model.

The Role of the Model Software

The model software should be used to guide the research. The model software should be used to document the model. The model software should be used to report the model.

The model software should be used to interpret the model. The model software should be used to acknowledge the model limitations. The model software should be used to promote the model.

The Role of the Model Code

The model code should be used to guide the research. The model code should be used to document the model. The model code should be used to report the model.

The model code should be used to interpret the model. The model code should be used to acknowledge the model limitations. The model code should be used to promote the model.

The Role of the Model Data

The model data should be used to guide the research. The model data should be used to document the model. The model data should be used to report the model.

The model data should be used to interpret the model. The model data should be used to acknowledge the model limitations. The model data should be used to promote the model.

The Role of the Model Output

The model output should be used to guide the research. The model output should be used to document the model. The model output should be used to report the model.

The model output should be used to interpret the model. The model output should be used to acknowledge the model limitations. The model output should be used to promote the model.

The Role of the Model Result

The model result should be used to guide the research. The model result should be used to document the model. The model result should be used to report the model.

The model result should be used to interpret the model. The model result should be used to acknowledge the model limitations. The model result should be used to promote the model.

The Role of the Model Finding

The model finding should be used to guide the research. The model finding should be used to document the model. The model finding should be used to report the model.

The model finding should be used to interpret the model. The model finding should be used to acknowledge the model limitations. The model finding should be used to promote the model.

The Role of the Model Conclusion

The model conclusion should be used to guide the research. The model conclusion should be used to document the model. The model conclusion should be used to report the model.

The model conclusion should be used to interpret the model. The model conclusion should be used to acknowledge the model limitations. The model conclusion should be used to promote the model.

The Role of the Model Recommendation

The model recommendation should be used to guide the research. The model recommendation should be used to document the model. The model recommendation should be used to report the model.

The model recommendation should be used to interpret the model. The model recommendation should be used to acknowledge the model limitations. The model recommendation should be used to promote the model.

The Role of the Model Suggestion

The model suggestion should be used to guide the research. The model suggestion should be used to document the model. The model suggestion should be used to report the model.

The model suggestion should be used to interpret the model. The model suggestion should be used to acknowledge the model limitations. The model suggestion should be used to promote the model.

The Role of the Model Advice

The model advice should be used to guide the research. The model advice should be used to document the model. The model advice should be used to report the model.

The model advice should be used to interpret the model. The model advice should be used to acknowledge the model limitations. The model advice should be used to promote the model.

The Role of the Model Guidance

The model guidance should be used to guide the research. The model guidance should be used to document the model. The model guidance should be used to report the model.

The model guidance should be used to interpret the model. The model guidance should be used to acknowledge the model limitations. The model guidance should be used to promote the model.

The Role of the Model Direction

The model direction should be used to guide the research. The model direction should be used to document the model. The model direction should be used to report the model.

The model direction should be used to interpret the model. The model direction should be used to acknowledge the model limitations. The model direction should be used to promote the model.

The Role of the Model Instruction

The model instruction should be used to guide the research. The model instruction should be used to document the model. The model instruction should be used to report the model.

The model instruction should be used to interpret the model. The model instruction should be used to acknowledge the model limitations. The model instruction should be used to promote the model.

The Role of the Model Specification

The model specification should be used to guide the research. The model specification should be used to document the model. The model specification should be used to report the model.

The model specification should be used to interpret the model. The model specification should be used to acknowledge the model limitations. The model specification should be used to promote the model.

The Role of the Model Description

The model description should be used to guide the research. The model description should be used to document the model. The model description should be used to report the model.

The model description should be used to interpret the model. The model description should be used to acknowledge the model limitations. The model description should be used to promote the model.

The Role of the Model Explanation

The model explanation should be used to guide the research. The model explanation should be used to document the model. The model explanation should be used to report the model.

The model explanation should be used to interpret the model. The model explanation should be used to acknowledge the model limitations. The model explanation should be used to promote the model.

The Role of the Model Clarification

The model clarification should be used to guide the research. The model clarification should be used to document the model. The model clarification should be used to report the model.

The model clarification should be used to interpret the model. The model clarification should be used to acknowledge the model limitations. The model clarification should be used to promote the model.

The Role of the Model Elaboration

The model elaboration should be used to guide the research. The model elaboration should be used to document the model. The model elaboration should be used to report the model.

The model elaboration should be used to interpret the model. The model elaboration should be used to acknowledge the model limitations. The model elaboration should be used to promote the model.

The Role of the Model Expansion

The model expansion should be used to guide the research. The model expansion should be used to document the model. The model expansion should be used to report the model.

The model expansion should be used to interpret the model. The model expansion should be used to acknowledge the model limitations. The model expansion should be used to promote the model.

The Role of the Model Extension

The model extension should be used to guide the research. The model extension should be used to document the model. The model extension should be used to report the model.

The model extension should be used to interpret the model. The model extension should be used to acknowledge the model limitations. The model extension should be used to promote the model.

The Role of the Model Generalization

The model generalization should be used to guide the research. The model generalization should be used to document the model. The model generalization should be used to report the model.

The model generalization should be used to interpret the model. The model generalization should be used to acknowledge the model limitations. The model generalization should be used to promote the model.

The Role of the Model Application

The model application should be used to guide the research. The model application should be used to document the model. The model application should be used to report the model.

The model application should be used to interpret the model. The model application should be used to acknowledge the model limitations. The model application should be used to promote the model.

The Role of the Model Implementation

The model implementation should be used to guide the research. The model implementation should be used to document the model. The model implementation should be used to report the model.

The model implementation should be used to interpret the model. The model implementation should be used to acknowledge the model limitations. The model implementation should be used to promote the model.

The Role of the Model Integration

The model integration should be used to guide the research. The model integration should be used to document the model. The model integration should be used to report the model.

The model integration should be used to interpret the model. The model integration should be used to acknowledge the model limitations. The model integration should be used to promote the model.

The Role of the Model Synthesis

The model synthesis should be used to guide the research. The model synthesis should be used to document the model. The model synthesis should be used to report the model.

The model synthesis should be used to interpret the model. The model synthesis should be used to acknowledge the model limitations. The model synthesis should be used to promote the model.

The Role of the Model Evaluation

The model evaluation should be used to guide the research. The model evaluation should be used to document the model. The model evaluation should be used to report the model.

The model evaluation should be used to interpret the model. The model evaluation should be used to acknowledge the model limitations. The model evaluation should be used to promote the model.

The Role of the Model Assessment

The model assessment should be used to guide the research. The model assessment should be used to document the model. The model assessment should be used to report the model.

The model assessment should be used to interpret the model. The model assessment should be used to acknowledge the model limitations. The model assessment should be used to promote the model.

The Role of the Model Review

The model review should be used to guide the research. The model review should be used to document the model. The model review should be used to report the model.

The model review should be used to interpret the model. The model review should be used to acknowledge the model limitations. The model review should be used to promote the model.

The Role of the Model Audit

The model audit should be used to guide the research. The model audit should be used to document the model. The model audit should be used to report the model.

The model audit should be used to interpret the model. The model audit should be used to acknowledge the model limitations. The model audit should be used to promote the model.

The Role of the Model Inspection

The model inspection should be used to guide the research. The model inspection should be used to document the model. The model inspection should be used to report the model.

The model inspection should be used to interpret the model. The model inspection should be used to acknowledge the model limitations. The model inspection should be used to promote the model.

The Role of the Model Examination

The model examination should be used to guide the research. The model examination should be used to document the model. The model examination should be used to report the model.

The model examination should be used to interpret the model. The model examination should be used to acknowledge the model limitations. The model examination should be used to promote the model.

The Role of the Model Investigation

The model investigation should be used to guide the research. The model investigation should be used to document the model. The model investigation should be used to report the model.

The model investigation should be used to interpret the model. The model investigation should be used to acknowledge the model limitations. The model investigation should be used to promote the model.

The Role of the Model Exploration

The model exploration should be used to guide the research. The model exploration should be used to document the model. The model exploration should be used to report the model.

The model exploration should be used to interpret the model. The model exploration should be used to acknowledge the model limitations. The model exploration should be used to promote the model.

The Role of the Model Discovery

The model discovery should be used to guide the research. The model discovery should be used to document the model. The model discovery should be used to report the model.

The model discovery should be used to interpret the model. The model discovery should be used to acknowledge the model limitations. The model discovery should be used to promote the model.

The Role of the Model Innovation

The model innovation should be used to guide the research. The model innovation should be used to document the model. The model innovation should be used to report the model.

The model innovation should be used to interpret the model. The model innovation should be used to acknowledge the model limitations. The model innovation should be used to promote the model.

The Role of the Model Creation

The model creation should be used to guide the research. The model creation should be used to document the model. The model creation should be used to report the model.

The model creation should be used to interpret the model. The model creation should be used to acknowledge the model limitations. The model creation should be used to promote the model.

The Role of the Model Development

The model development should be used to guide the research. The model development should be used to document the model. The model development should be used to report the model.

The model development should be used to interpret the model. The model development should be used to acknowledge the model limitations. The model development should be used to promote the model.

The Role of the Model Construction

The model construction should be used to guide the research. The model construction should be used to document the model. The model construction should be used to report the model.

The model construction should be used to interpret the model. The model construction should be used to acknowledge the model limitations. The model construction should be used to promote the model.

The Role of the Model Building

The model building should be used to guide the research. The model building should be used to document the model. The model building should be used to report the model.

The model building should be used to interpret the model. The model building should be used to acknowledge the model limitations. The model building should be used to promote the model.

The Role of the Model Fitting

The model fitting should be used to guide the research. The model fitting should be used to document the model. The model fitting should be used to report the model.

The model fitting should be used to interpret the model. The model fitting should be used to acknowledge the model limitations. The model fitting should be used to promote the model.

The Role of the Model Estimation

The model estimation should be used to guide the research. The model estimation should be used to document the model. The model estimation should be used to report the model.

The model estimation should be used to interpret the model. The model estimation should be used to acknowledge the model limitations. The model estimation should be used to promote the model.

The Role of the Model Prediction

The model prediction should be used to guide the research. The model prediction should be used to document the model. The model prediction should be used to report the model.

The model prediction should be used to interpret the model. The model prediction should be used to acknowledge the model limitations. The model prediction should be used to promote the model.

The Role of the Model Forecasting

The model forecasting should be used to guide the research. The model forecasting should be used to document the model. The model forecasting should be used to report the model.

The model forecasting should be used to interpret the model. The model forecasting should be used to acknowledge the model limitations. The model forecasting should be used to promote the model.

The Role of the Model Projection

The model projection should be used to guide the research. The model projection should be used to document the model. The model projection should be used to report the model.

The model projection should

Frequently Asked Questions

What is the main difference between Cox and parametric survival models?

The Cox model leaves the baseline hazard unspecified, while parametric models assume a specific distribution for the survival times. The Cox model estimates hazard ratios without specifying the baseline hazard, while parametric models estimate the full survival distribution.

When should I use a Cox proportional hazards model?

Use the Cox model when you do not know the underlying survival distribution and want to estimate hazard ratios for covariates. The Cox model is robust to the distribution of the baseline hazard and is the default choice for many survival analyses.

When should I use a parametric survival model?

Use a parametric model when you need to predict survival times or survival probabilities, or when you have a strong theoretical basis for the distribution of survival times. Parametric models are more efficient than the Cox model when the distribution is correct.

How do I test the proportional hazards assumption?

The proportional hazards assumption can be tested using the Schoenfeld residuals. A significant correlation between the residuals and time indicates a violation of the assumption. Graphical methods, such as plotting the log cumulative hazard, can also be used.

What is the AIC and how is it used in model selection?

The Akaike Information Criterion (AIC) is a measure of the relative quality of a statistical model. The AIC is calculated as the negative log-likelihood plus a penalty for the number of parameters. Lower AIC values indicate a better fit.

Can I compare the Cox model to a parametric model?

The Cox model and the parametric models can be compared using the AIC if the parametric models are proportional hazards models. The Cox model has a partial likelihood that can be compared to the full likelihood of the parametric models.

What are the limitations of the Cox model?

The Cox model does not provide a full survival distribution, so it cannot be used to predict survival times. The Cox model also relies on the proportional hazards assumption, which may not hold in all cases.

What are the limitations of parametric models?

Parametric models rely on the distributional assumption. If the distribution is incorrect, the model is misspecified and the estimates are biased. The parametric models are also less flexible than the Cox model.

Related Bioinformatics Guides

• Genomic Data Analysis Tools: A Comparative Guide for Researchers
• Persistent Identifiers for Research Data: A Guide to Selection and Use
• Spatial Transcriptomics vs. Single-Cell RNA Sequencing: Which Approach Fits Your Research?
• Data Stewardship vs Data Governance: What's the Difference?
• Foundation Models in Genetics: Opportunities and Challenges

Related Clinical & Scientific Guides

• A Practical Guide to Detecting Antimicrobial Resistance Genes in Shotgun Metagenomic Data
• Computational Immunology: Modeling the Immune System
• How to Set Hard Filters for Germline Variant Calling: A Practical Guide to GATK Best Practices

References and Further Reading

This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.