Quick Definition: A longitudinal study is a research design in which the same individuals are observed or measured repeatedly over an extended period of time (from weeks to decades) to track changes, identify patterns, and establish cause-and-effect relationships between variables. Longitudinal studies can be experimental or observational.
What is a Longitudinal Study?
A longitudinal study is a type of study in which the same group of individuals (called a cohort) is followed and measured repeatedly over a defined period of time. The defining feature is continuity: the same subjects are tracked through time, rather than different groups being observed at a single moment.
These studies are can be either observational (no intervention or manipulation of variables by the researcher) or experimental. In an experimental longitudinal study, the researcher actively intervenes (e.g., assigning participants to a treatment or programme) and then tracks outcomes in those same individuals repeatedly over time. A long-term clinical trial measuring patient outcomes at six months, one year, and two years post-treatment is a straightforward example of an experimental longitudinal design.
In practice, however, the majority of longitudinal research is observational, because many of the questions best suited to this design cannot be studied through deliberate intervention (e.g., how does a risk factor relate to disease development over decades? how do childhood experiences shape adult health?).
Longitudinal research is particularly valuable for:
- Evaluating the relationship between risk factors and the development of disease
- Tracking treatment outcomes over different time horizons
- Understanding how individual-level characteristics change across a lifespan
- Establishing the correct sequence of events so that causality can be more confidently inferred
The duration of a longitudinal study has no fixed minimum. Studies can span weeks, years, or generations. What matters is that the same participants are observed more than once. In practice, most longitudinal studies run for at least a year, and many run for decades.
Types of Longitudinal Study Designs
Longitudinal research takes several distinct forms, each suited to different research questions and resource constraints.
1. Prospective Studies
In a prospective longitudinal study, participants are recruited before the outcomes of interest have occurred. Researchers then follow them forward in time, collecting data as events unfold. This is the “gold standard” form of longitudinal research because it allows for careful, pre-planned data collection and eliminates recall bias.
Example:
A group of 1,000 non-smokers is enrolled and monitored over 30 years to observe the incidence of lung disease.
Prospective studies include several sub-types based on how the sample is defined:
- Cohort panels: all or some individuals in a defined population sharing similar characteristics (e.g., born in the same year, exposed to the same occupational hazard) are observed over time.
- Representative panels: data are regularly collected from a random sample of a broader population, enabling generalisable findings.
- Linked panels: data originally collected for other purposes (e.g., administrative or health records) are linked together to create individual-level datasets.
2. Retrospective Studies
In a retrospective longitudinal study, the research is designed after some or all of the relevant events have already occurred. Researchers look backwards, examining existing records, such as medical histories, national registers, or archives, to identify exposures and outcomes.
Retrospective studies are generally less expensive and faster than prospective designs, but they are more vulnerable to measurement error and are constrained by whatever data happened to be recorded at the time.
Example:
Examining the medical records of patients over the past 20 years to identify which early risk factors are associated with a specific disease.
A well-known example is the research into the fetal origins of coronary heart disease, in which researchers traced groups of men and women born in Hertfordshire before 1930 whose fetal and infant growth had been documented. By linking historical birth records to death certificates, they could relate death rates from coronary heart disease to birth weight and weight at one year old.
3. Repeated Cross-Sectional Studies
In this design, data are collected from different (mostly or entirely new) participants at each measurement occasion, rather than following the same individuals. This allows researchers to track changes at the population level over time, but it cannot capture individual-level change or development.
Example:
Annual surveys of voter intentions using different randomly selected samples each year.
4. Clinical Follow-Up Studies
A specialised form of longitudinal research used extensively in medicine, clinical follow-up studies systematically monitor patients with a particular disease or condition to establish how their illness progresses and what factors influence prognosis.
Clinical impressions alone are often misleading. A specialist’s view of a disease may be skewed toward severe cases, while a general practitioner sees too few patients to form a reliable picture. Systematic longitudinal follow-up corrects for this. Outcomes in clinical studies are often expressed as case fatality rates or survival rates, which can be plotted over time to produce survival curves.
Longitudinal Study vs Cross-Sectional Study
The most common comparison is between longitudinal and cross-sectional studies. Both are generally observational (researchers do not manipulate variables) but they differ substantially in structure and what they can reveal.
A cross-sectional study collects data from a population at a single point in time. It provides a snapshot of a group or society at that moment, and is useful for identifying the prevalence of a condition or examining associations between variables as they exist at one moment. However, because time is not a factor, cross-sectional designs cannot establish the sequence of events and are therefore weaker tools for inferring causality.
A longitudinal study, by contrast, follows the same individuals across multiple time points. This allows researchers to observe change within individuals, establish what came before what, and build a stronger case for causal relationships.
| Characteristic | Longitudinal Study | Cross-Sectional Study |
| Observation | Same individuals, multiple time points | Different (or same) individuals, single time point |
| Primary use | Tracking change over time; causality | Describing prevalence; snapshot associations |
| Causality | Stronger basis for causal inference | Cannot establish sequence of events |
| Duration | Weeks to decades | Short (data collected at one moment) |
| Cost | Higher, due to extended follow-up | Lower |
| Recall bias | Prospective designs eliminate recall bias | More susceptible to recall bias |
| Attrition | A major challenge | Not applicable |
| Example | Tracking cardiovascular risk factors in 5,000 adults over 20 years | Surveying smoking prevalence in a community at one point in time |
A useful workflow is to conduct a cross-sectional study first to identify whether an association worth investigating exists, and then design a longitudinal study to examine that association in depth. For example, a cross-sectional study might reveal a link between a dietary pattern and a health outcome in men but not women. This is a finding that could then guide the design of a targeted longitudinal study in men.
Longitudinal Study vs Other Research Designs
Longitudinal vs Cohort Study
The terms are often used interchangeably, but there is a technical distinction. A cohort study is a type of longitudinal study that specifically compares groups (cohorts) defined by a shared characteristic (often exposure to a risk factor) and follows them over time. All cohort studies are longitudinal, but not all longitudinal studies are strictly cohort studies.
Longitudinal vs Case Study
A case study is an in-depth investigation of a single individual, group, or event at a particular time. Longitudinal research, by contrast, involves repeated measurement of the same participants over time and typically aims to produce generalisable findings across a population.
Longitudinal vs Experimental Study
Experimental studies (such as randomised controlled trials) involve active manipulation of an independent variable and random assignment of participants to conditions. Longitudinal studies are mostly observational: variables are measured, not manipulated. Experimental designs generally offer stronger internal validity for establishing causality, but longitudinal studies are often more practical and ethically feasible when studying natural exposures over long timeframes.
How to Conduct a Longitudinal Study
Conducting a longitudinal study is demanding because the infrastructure must remain robust for the entire duration of the study. The following steps outline the core process.
Step 1: Define Your Research Question and Objectives
Identify the specific changes or relationships you want to investigate, so that you can formulate your research question and research objectives. Clarify your primary endpoints: are you primarily interested in absolute outcomes (does disease develop or not?) or in variation over time (how does a biomarker change from year to year)? The answer will shape every subsequent decision: the study duration, sampling frequency, and statistical approach.
Step 2: Choose Your Study Design
Decide between a prospective or retrospective approach based on your question, timeline, and budget. Consider whether existing datasets like government surveys, health registries, or cohort data made available through repositories, might meet your needs before committing to expensive primary data collection.
Step 3: Select and Recruit Your Sample
Your sample should be representative of the population you wish to draw conclusions about, and large enough to ensure statistical power for your primary endpoints. Plan from the outset for attrition: some proportion of participants will drop out, move away, become ill, or die. A sample that adequately represents your population at the start may not do so five years in, which will threaten the validity of your conclusions. Strategies for maximising retention should be built into the study protocol from the beginning.
Step 4: Standardise Data Collection Methods
This is critical and non-negotiable. Data collection and recording methods must be identical across all study sites and must remain consistent over time. All data should be classified according to the measurement interval, and every data point must be linked to the correct individual through a unique coding system. Using recognised, validated classification systems for individual inputs improves accuracy and comparability.
Step 5: Develop a Data Management Plan
Longitudinal studies generate large volumes of data over many years. Establish a plan for secure storage, access, version control, and ongoing quality checks before data collection begins. Consider how data from different time points will be linked and how missing data will be handled.
Step 6: Set the Sampling Frequency
How often you collect data should depend on the nature of your endpoints. Outcomes that change slowly (e.g., bone density, cognitive function) may only need annual or biennial measurement. Rapidly changing outcomes (e.g., lung function in firefighters exposed to fumes) may require daily or weekly monitoring.
Step 7: Maintain Participant Engagement
Attrition is the Achilles’ heel of longitudinal research. Implement retention strategies systematically:
- Involve the community in study design and recruitment
- Explain objectives, benefits, and risks transparently to participants
- Send regular reminders about appointments or data collection
- Offer financial reimbursement for research-related expenses and non-financial tokens of appreciation
- Create convenient methods for remote follow-up so that geographic mobility does not automatically lead to dropout
- Conduct exit interviews with participants who do leave, to understand the reasons
Step 8: Monitor, Adapt, and Analyse
Regular monitoring of outcome measures and focused review of any emerging concerns is essential. Longitudinal studies are dynamic. Procedures may need updating and contributors may need retraining as circumstances change. At the analysis stage, use statistical methods appropriate for repeated-measures data.
Statistical Analysis in Longitudinal Research
Analysing longitudinal data requires methods that account for the fact that repeated observations from the same individual are not independent of each other. Using standard cross-sectional statistical approaches on longitudinal data is a common error. It leads to underestimation of variability and an increased risk of false-negative results (Type II error).
Key factors to consider when selecting a statistical approach include:
- The linked nature of data for each individual despite separation in time
- The co-existence of fixed variables (e.g., sex) and dynamic variables (e.g., blood pressure)
- Potential differences in time intervals between data collection points
- The likely presence of missing data
The three most commonly applied approaches are:
Analysis of Variance (ANOVA/MANOVA)
Univariate (ANOVA) and multivariate (MANOVA) analysis of variance are often used for longitudinal analysis. These methods assume equal interval lengths between measurements and normally distributed data. A limitation is that they compare means only, sacrificing individual-level data.
2. Mixed-Effect Regression Models (MRM)
Mixed-effect models are better suited to longitudinal data because they focus specifically on individual change over time, account for variation in the timing of repeated measures, and can handle missing or unequal data instances. They are widely used in medical and psychological longitudinal research.
3. Generalised Estimating Equations (GEE)
GEE models rely on the independence of individuals within the population (rather than within-person correlation) and focus primarily on regression-level trends across the population. They are useful when population-average effects are of more interest than individual trajectories.
Advantages of Longitudinal Studies
Establish the correct sequence of events
By observing the same individuals over time, researchers can determine what preceded what. This is a prerequisite for causal inference. A cross-sectional study might find that higher police presence correlates with higher crime rates, wrongly implying causation; a longitudinal study would reveal the direction of the relationship.
Track change within individuals
Longitudinal designs allow researchers to observe how outcomes evolve within the same person, removing the confounding effect of individual differences. In a study of a weight-training programme, for example, the effect of natural athletic talent on performance is controlled for because it remains constant within each individual across the study period.
Eliminate recall bias
Prospective longitudinal studies collect data in real time, before participants know what the outcomes will be. This eliminates the risk of recall bias: the tendency for people to misremember past events, especially in light of what happened later.
Account for the cohort effect
A sophisticated advantage of longitudinal designs is the ability to separate and analyse the distinct contributions of cohort (range of birth dates), period (the current point in historical time), and age (how old the individual is at measurement). This is known as correcting for the “cohort effect” and is impossible with cross-sectional data.
High internal validity
Because objectives, variables, and measurement protocols are established before data collection begins, longitudinal studies (particularly prospective ones) tend to have high internal validity.
Flexibility for follow-up investigation
Patterns observed mid-study can prompt new hypotheses. Because the infrastructure and the cohort are already in place, researchers can investigate new questions without starting from scratch.
Disadvantages of Longitudinal Studies
Time-consuming and expensive
The extended duration of longitudinal studies requires sustained funding, stable research teams, and ongoing participant engagement, all of which are costly. This is one of the main barriers to conducting high-quality longitudinal research.
Attrition and selective dropout
Participants drop out of longitudinal studies for many reasons: illness, relocation, disinterest, or death. If the dropout is random, the main consequence is reduced sample size. More dangerously, if dropout is related to the exposure or outcome being studied (known as selective attrition), the remaining sample may no longer be representative, threatening the validity of the conclusions.
For example, in a study on the effectiveness of a diet programme, participants who are not seeing results may be more likely to drop out, thus making the programme appear more successful than it actually is.
Report bias
Longitudinal studies that rely on self-reported data (surveys, questionnaires, diaries) are vulnerable to report bias, since there is no independent way to verify participant responses.
Difficulty separating reciprocal effects
When two variables influence each other over time, it becomes challenging to cleanly separate cause from effect, especially when the induction period between exposure and outcome is long. For example, depression affects physical activity and physical activity also affects depression.
Statistical complexity
The repeated-measures nature of the data requires more sophisticated statistical techniques than cross-sectional analysis. Choosing inappropriate methods (such as applying standard ANOVA as though each time point were an independent sample) is a common and consequential analytical error.
Risk of historical confounding
Over a long study period, external factors like changes in diet, medical technology, social policy, or the environment may influence outcomes in ways that are difficult to separate from the study variables themselves.
Famous Longitudinal Study Examples
1. The Framingham Heart Study (1948–present)
Perhaps the most cited longitudinal study in medical history. Beginning in 1948, researchers recruited 5,209 adults aged 30–62 from Framingham, Massachusetts, and followed them, and subsequently their children and grandchildren. The study identified the major risk factors for cardiovascular disease (CVD), including hypertension, high cholesterol, smoking, physical inactivity, and obesity.
The absence of effective treatments such as statins and antihypertensives in the early decades of the study allowed researchers to observe the natural history of CVD with unusual clarity. The study’s generalisability has been questioned given its origin in a single town, though Framingham’s demographic diversity helped mitigate this concern.
2. The Harvard Study of Adult Development (1938–present)
One of the longest-running studies of adult life, following the same group of Boston men and later their families for more than 80 years. The study examines the physical, psychological, and social factors that predict healthy ageing and wellbeing, with a particular focus on the role of relationships and childhood experiences on midlife health.
3. The 1970 British Cohort Study (1970–present)
This study has tracked the lives of approximately 17,000 people born in England, Scotland, and Wales during a single week in April 1970. It collects data on health, education, economic circumstances, employment, family life, and social attitudes. The raw data are freely available through the UK Data Service, making it a valuable resource for secondary researchers.
4. The Office of Population Censuses and Surveys (OPCS) Longitudinal Study (1971–present)
This study prospectively follows a 1% sample of the British population identified in the 1971 census. It has linked census data at successive time points with vital event records, enabling researchers to investigate how employment status, housing, and other socioeconomic factors relate to mortality and the incidence of cancer and other diseases over time.
5. Hertfordshire Cohort Study (Retrospective)
To investigate the fetal origins of coronary heart disease, researchers identified groups of men and women born in Hertfordshire before 1930 whose fetal and infant growth had been documented. By tracing these individuals and linking birth records to death certificates, researchers could examine how birth weight and weight at one year of age predicted death rates from coronary heart disease decades later. This is a landmark retrospective longitudinal study.
6. Occupational Exposure to Ethylene Oxide (Cohort Study)
In a cohort study of workers occupationally exposed to ethylene oxide (a sterilant gas and antifreeze precursor), national death rates served as the reference population rather than a separately recruited control group, since general population exposure was negligible. The number of deaths in the cohort was compared against the expected number based on age-, sex-, and calendar-period-specific national rates, a method known as calculating standardised mortality ratios.
Longitudinal Studies in Different Fields
Medicine and Epidemiology
Longitudinal research is the backbone of chronic disease epidemiology. Studies tracking cardiovascular risk, cancer incidence, diabetes progression, and the long-term outcomes of surgical and pharmaceutical interventions are central to evidence-based medicine. Clinical follow-up studies are also used extensively in prognosis research, establishing survival rates and identifying predictors of disease course.
Psychology and Mental Health
Longitudinal studies in psychology have fundamentally shaped our understanding of human development across the lifespan. They have tracked how early childhood experiences influence adult mental health, how cognitive function changes with age, and how factors such as social support, trauma, and personality traits predict long-term outcomes.
Education and Social Sciences
Educational longitudinal studies follow cohorts of students from early childhood through adulthood, examining how school environments, family background, and socioeconomic factors influence educational attainment, employment, and life outcomes.
Economics and Policy Research
Economists and policy researchers use panel data, a form of longitudinal data, to evaluate the effects of policy interventions over time. Annual surveys tracking household income, employment, and consumption across the same households are a staple of economic longitudinal research.
Ethical Considerations in Longitudinal Studies
Longitudinal studies carry specific ethical responsibilities that go beyond those of shorter research projects.
Informed consent over time
Participants must provide informed consent at enrolment, but for very long studies, especially those that begin in childhood, consent processes may need to be revisited as participants age, circumstances change, or new research questions emerge.
Data privacy and security
Maintaining identifiable longitudinal records over decades requires robust data protection protocols, particularly as data are linked across time points and potentially across datasets.
Equitable burden-sharing
Participants in long longitudinal studies may be asked to give substantial time over many years. Ensuring fair compensation and minimising burden are important ethical obligations.
Exit and attrition
Participants must remain free to withdraw at any time. Exit interviews should be conducted sensitively, and the reasons for dropout should be documented to allow bias assessment without violating the confidentiality of those who withdraw.
Feedback to participants
Where clinically relevant findings emerge during the study, researchers have an ethical obligation to consider how and whether to communicate these findings to participants.
Biases in Longitudinal Studies
Major Bias Types & How to Address Them
| Bias | What It Is | How to Address It |
| Attrition (Drop-out) Bias | Participants who leave differ systematically from those who stay (e.g., sicker people drop out of health studies) | Intention-to-treat analysis; multiple imputation; inverse probability weighting; collect reasons for dropout |
| Survivorship Bias | Only “survivors” remain in the sample over time, skewing results optimistically | Carefully track all dropouts; sensitivity analyses including worst-case scenarios for missing data |
| Cohort Effect | Findings reflect the specific historical/cultural experience of one generation, not universal patterns | Use multiple cohorts born at different times; age-period-cohort models |
| Testing / Practice Effect | Repeated exposure to the same measures improves performance artificially (common in cognitive studies) | Alternate test forms; include control groups; model practice effects statistically |
| Regression to the Mean | Extreme measurements at baseline tend to move toward the average over time | Include a control group; use statistical methods that account for baseline values |
| Temporal Ambiguity | Even with repeated measures, the exact timing of cause and effect can be unclear | Cross-lagged panel models; fine-grained measurement intervals; diary or ecological momentary assessment |
| Historical / Period Effects | External events (wars, pandemics, recessions) affect all participants at the same time, confounding results | Multi-cohort designs; clearly document historical context; sensitivity analyses around event windows |
| Instrumentation Drift | Measurement tools, raters, or protocols change over time, introducing artifactual trends | Standardize protocols; calibrate instruments regularly; use anchor items across waves |
| Selective Recruitment Bias | The initial sample isn’t representative (e.g., volunteers tend to be healthier, wealthier) | Random sampling; oversampling underrepresented groups; report and model baseline characteristics |
| Maturation Effects | Participants naturally change with age, which can be mistaken for a treatment or exposure effect | Include a comparison group; use age as a covariate; distinguish developmental from intervention effects |
General Best Practices
- Pre-register hypotheses and analysis plans to reduce fishing for results across waves
- Maximize retention with reminders, incentives, and flexible contact methods — and track why people drop out
- Use mixed models / GEE (Generalized Estimating Equations) which handle missing data and repeated measures more robustly than naive repeated ANOVA
- Report dropout rates and compare completers vs. non-completers on key baseline variables
- Sensitivity analyses: re-run key analyses under different assumptions about missing data to test how robust findings are
Key Takeaways
- A longitudinal study tracks the same individuals repeatedly over time to observe change, detect patterns, and investigate cause-and-effect relationships.
- The three main designs are prospective studies (following participants forward in time), retrospective studies (using existing records to look backwards), and repeated cross-sectional studies (different participants at each time point, tracking population-level change).
- Longitudinal studies are more powerful than cross-sectional studies for establishing causality because they can confirm the sequence of events, but they are costlier and logistically more complex.
- The biggest methodological challenges are attrition (participant dropout), report bias (from self-reported data), and statistical complexity (requiring methods like mixed-effect regression or GEE that account for within-individual correlation).
- Landmark longitudinal studies like the Framingham Heart Study and the Harvard Study of Adult Development have produced some of the most important and durable findings in modern science.
- Longitudinal research is used across medicine, epidemiology, psychology, education, economics, and social science.
Frequently Asked Questions
What is the difference between a longitudinal study and a cross-sectional study?
A longitudinal study follows the same individuals over multiple time points, allowing researchers to track change and establish the sequence of events. A cross-sectional study collects data from a population at a single moment in time, providing a snapshot but no information about how variables change. Longitudinal studies are better for causal inference; cross-sectional studies are cheaper and faster.
Are longitudinal studies qualitative or quantitative?
Longitudinal studies can be either, or both. Quantitative longitudinal research uses numerical data and statistical analysis to identify trends and causal relationships. This is common in medicine, economics, and psychology. Qualitative longitudinal research tracks how individuals’ experiences, narratives, and perspectives evolve over time through methods like repeated interviews or observations. This common in sociology and qualitative health research.
What is the typical length of a longitudinal study?
There is no fixed requirement. Studies can range from a few weeks (e.g., monitoring acute effects of occupational exposure) to several decades (e.g., the Framingham Heart Study, now in its eighth decade). Most aim to last at least a year to capture meaningful change over time.
What is selective attrition, and why does it matter?
Selective attrition refers to dropout that is related to the variables being studied. For example, if sicker participants are more likely to leave a health study, the remaining sample will be unrepresentatively healthy, leading to biased conclusions. It is one of the most serious threats to the validity of longitudinal research.
What statistical methods are used in longitudinal studies?
Common approaches include mixed-effect regression models (MRM), which track individual change over time, and generalised estimating equations (GEE), which focus on population-average effects. Standard cross-sectional methods such as basic ANOVA should not be applied to repeated-measures data without modification, as they underestimate variability and increase the risk of false negatives.
What is the cohort effect, and how do longitudinal studies handle it?
The cohort effect refers to the influence that the historical period during which a person was born or grew up has on their outcomes, distinct from the effects of their current age or the current time period. Longitudinal studies can separate and account for these three time components (cohort, period, and age), which cross-sectional studies cannot.
How do researchers handle participant dropout in longitudinal studies?
Retention strategies include transparent communication about study objectives and benefits, regular appointment reminders, financial reimbursements, convenient remote follow-up options, and community involvement in study design. Exit interviews with participants who leave provide valuable information about potential attrition bias.
Can I access longitudinal study data without conducting my own study?
Yes. Many governments and research institutions publish longitudinal datasets freely. For example, data from the 1970 British Cohort Study are available through the UK Data Service. Using existing data is less expensive and can allow analysis over longer timeframes, though researchers are limited to the variables and formats defined by the original study team.
What is intention-to-treat analysis?
The core idea of intention-to-treat analysis is that you analyze participants in the group they were randomly assigned to, regardless of whether they actually complied with or completed the treatment.
Why It Matters
- Prevents selection bias: non-compliers are rarely random; excluding them distorts results
- Preserves the benefits of randomization (group balance on known and unknown confounders)
- Reflects real-world effectiveness, not just ideal-condition efficacy
How It’s Performed
| Step | Action |
| 1. Randomize | Assign all participants to groups at baseline |
| 2. Keep everyone in | Retain all randomized participants in their assigned group for analysis, even if they: switched groups, dropped out, never started treatment, or violated protocol |
| 3. Handle missing outcome data | Use one of the methods below (this is the hardest part) |
| 4. Analyze by assigned group | Run the primary analysis (e.g., t-test, regression) on the full assigned sample |
Handling Missing Data (the Critical Challenge)
- Complete case ITT: only analyze those with outcome data; simple but risks bias if dropout is non-random
- Last observation carried forward (LOCF): impute missing values with the participant’s last known value; easy but often unrealistic. We don’t recommend this method.
- Multiple imputation (MI): statistically generates plausible missing values many times based on observed data, then pools results; widely recommended
- Mixed models for repeated measures (MMRM): models all available time-point data without explicitly imputing; preferred in clinical trials
- Worst-case / best-case sensitivity analyses: test how robust conclusions are under extreme assumptions about missing data
ITT vs. Related Approaches
| Approach | Who Is Analyzed | Use Case |
| ITT | All randomized participants | Primary efficacy/effectiveness analysis |
| Per-protocol (PP) | Only those who fully complied | Supplementary; shows biological efficacy under ideal conditions |
| Modified ITT (mITT) | All randomized except clearly ineligible (e.g., never received any treatment) | Common compromise; must be pre-specified |
Key Takeaway
ITT is conservative in that it typically underestimates treatment effects. But it gives the most honest, bias-resistant estimate of how a treatment performs when deployed in the real world.
References
- Caruana EJ, Roman M, Hernández-Sánchez J, Solli P. Longitudinal studies. J Thorac Dis. 2015;7(11):E537–E540. doi:10.3978/j.issn.2072-1439.2015.10.63.
- Dawber TR, Kannel WB, Lyell LP. An approach to longitudinal studies in a community: the Framingham Study. Ann N Y Acad Sci. 1963;107:539–556.
- Doll D. Chapter 7: Longitudinal studies. In: Epidemiology for the Uninitiated. BMJ Publishing Group. Available at: https://www.bmj.com/about-bmj/resources-readers/publications/epidemiology-uninitiated/7-longitudinal-studies
- Thomas L. Longitudinal Study | Definition, Approaches & Examples. Scribbr. Updated June 22, 2023. https://www.scribbr.com/methodology/longitudinal-study/
- Tooth L, Ware R, Bain C, et al. Quality of reporting of observational longitudinal research. Am J Epidemiol. 2005;161:280–288.
- Edwards LJ. Modern statistical techniques for the analysis of longitudinal data in biomedical research. Pediatr Pulmonol. 2000;30:330–344.
- Newman AB. An overview of the design, implementation, and analyses of longitudinal studies on aging. J Am Geriatr Soc. 2010;58(Suppl 2):S287–291.
- Abshire M et al. Participant retention practices in longitudinal clinical research studies with high retention rates. BMC Med Res Methodol. 2017;17(1):30.
This article was originally published on January 8, 2025, and updated on June 09, 2026.
