Formulating a Research Question

Transcript

• 00:02<v Maria>My name is Maria Ciarleglio</v>
• 00:04and I'm a faculty member
• 00:05in the Department of Biostatistics
• 00:08at the Yale School of Public Health.
• 00:11In this video series I will introduce the clinical research
• 00:14process to prepare you to collaborate with a statistician.
• 00:20In this first video we'll discuss what is often
• 00:24the first step of the research process,
• 00:26formulating a research question.
• 00:31The first step in the research process
• 00:33is to convert the need for information
• 00:36into an answerable question or hypothesis.
• 00:40A well formulated research question is specific and precise.
• 00:45The research question guides the study design
• 00:49and other design-related study characteristics,
• 00:52the data that are collected, the data analysis
• 00:56and ultimately determines what you can conclude
• 00:59at the end of the study.
• 01:02The PICO criteria can be used to guide you
• 01:05in framing a comparative research question.
• 01:09The PICO framework begins by specifying the population
• 01:12of interest, then the intervention being studied,
• 01:16the control or comparator group,
• 01:19and the outcomes of interest.
• 01:23Begin by specifying the population of interest.
• 01:26For example,
• 01:27patients with non-alcoholic fatty liver disease.
• 01:31The target population is the group of patients
• 01:34to which you would like to generalize your study findings.
• 01:38The study population is the group
• 01:40of patients to which you have access.
• 01:44The study population may be a subset
• 01:47of the target population.
• 01:50For example, your goal may be to generalize
• 01:53to all adult Americans
• 01:55with non-alcoholic fatty liver disease.
• 01:58However, you may be limited to a patient population
• 02:02from a certain state or medical center.
• 02:06In our case, we may only have access to patients
• 02:09with non-alcoholic fatty liver disease
• 02:12followed in the liver clinic from 2015 to 2020.
• 02:18In this case, you could either collect data
• 02:20from all individuals in the available study population
• 02:24if it's feasible to do that.
• 02:26Otherwise, if the study population is too large
• 02:29you could select a random sample
• 02:32from that available study population.
• 02:35If you choose a representative random sample
• 02:39your results are generalizable to that study population.
• 02:45Next, specify the main intervention,
• 02:49which is the exposure test treatment
• 02:52or the main prognostic factor
• 02:54that you are interested in studying.
• 02:57For example, lifestyle modification to achieve weight loss
• 03:02or if studying liver cancer,
• 03:04your intervention of interest could be serafenib
• 03:07to prolonged survival.
• 03:12If you're interested in performing a comparison,
• 03:15the next step is to specify a control
• 03:18or comparison intervention or exposure.
• 03:23This can be, for example,
• 03:24a placebo control or the current standard of care.
• 03:30Finally, we must specify the clinical outcome
• 03:33or primary endpoint of your study.
• 03:37This includes the element of time, if that's appropriate,
• 03:40and this would apply if you're looking
• 03:42at a fixed follow up time period post-intervention.
• 03:47Say three month survival following surgery
• 03:51or NAFLD resolution one year following
• 03:54a certain percentage reduction in total body weight.
• 04:01Let's run through an example of the type of study
• 04:03we often perform using medical record data.
• 04:07The research question asks
• 04:10among Hepatitis B infected persons,
• 04:12what factors tests best identify individuals
• 04:17at highest risk of progression,
• 04:19as well as those at low risk of progression?
• 04:23The population studied is Hepatitis B infected persons
• 04:28treated at the Yale Liver Center between 2011 and 2021.
• 04:35The interventions of interest
• 04:37are different patient characteristics.
• 04:39Specifically, the study will look at different permutations
• 04:43of key baseline exposures or risk factors
• 04:47identified in previous studies of Hepatitis B prognosis.
• 04:52Here, the investigators will look at age
• 04:55presence of fibrosis, presence of cirrhosis,
• 04:58elevated ALT, and detectable viral load.
• 05:05The comparator group for each of these factors
• 05:08is absence of the baseline factor.
• 05:12The outcomes of interest are liver related morbidity,
• 05:16progression of liver disease
• 05:17and mortality during up to 10 years of follow up.
• 05:22Now, this is more of an exploratory study
• 05:25looking for signals of association, but even still,
• 05:29it has a clearly defined population,
• 05:31intervention or exposures of interest,
• 05:34control or reference levels of the exposures
• 05:37and outcomes of interest.
• 05:40Sitting down and thinking through the PICO criteria
• 05:43forces you to make decisions
• 05:45and pre-specify important aspects of your study.
• 05:51As we saw in the last example,
• 05:53there are often multiple clinical endpoints of interest.
• 05:57Endpoints are classified as clinical or nonclinical.
• 06:03Clinical endpoints describe outcomes
• 06:05involving how a patient feels, functions or survives.
• 06:10They may be assessed by a clinician
• 06:12and involve clinical judgment,
• 06:14such as the occurrence of stroke or MI.
• 06:17They may also be measured by a standard performance measure
• 06:21such as a pulmonary function test
• 06:23or they can be patient-reported,
• 06:25such as self-reported symptoms or quality of life.
• 06:30Nonclinical endpoints include biomarkers
• 06:33that may not directly relate to how a patient feels,
• 06:37however they're thought to be important indicators
• 06:39of the disease process.
• 06:41These endpoints can include blood tests, imaging
• 06:45or other physiological measures such as blood pressure.
• 06:49A good primary outcome should directly align
• 06:52with the primary aim of the study.
• 06:55The endpoint should be accurate
• 06:57and precise, quantifiable, validated, and reproducible.
• 07:02We generally include a single primary endpoint.
• 07:06The goal should be to choose a primary endpoint
• 07:09that will influence decision making in practice.
• 07:13The most significant and impactful endpoint that addresses
• 07:17the research question is chosen as the primary endpoint
• 07:21and additional important endpoints may be designated
• 07:25as secondary or tertiary.
• 07:28Secondary endpoints may not be considered sufficient
• 07:32to influence decision making alone,
• 07:35but may help support the claim of efficacy.
• 07:38Tertiary endpoints
• 07:39are sometimes called exploratory endpoints.
• 07:43If included, they are generally used
• 07:45to test exploratory hypotheses.
• 07:50Again, we generally use a single primary outcome.
• 07:54Using multiple primary endpoints may lead
• 07:57to an unfocused research question and can present problems
• 08:01with interpretation if the treatment effect is observed
• 08:04to differ across the multiple outcomes.
• 08:08However, multiple endpoints may be needed
• 08:11when a clinical benefit depends
• 08:13on more than one aspect of the disease.
• 08:16For example, in Alzheimer's, we may require an effect
• 08:20on both cognition and function,
• 08:23so there may be situations where multiple endpoints
• 08:26are necessary for demonstrating efficacy.
• 08:30The statistical issue with multiple endpoints
• 08:32is what we call multiplicity.
• 08:36When we conduct statistical analysis
• 08:38and perform hypothesis tests,
• 08:40there's a chance that we conclude
• 08:42a significant difference exists between the arms tested
• 08:47when in truth, there is no difference.
• 08:49This is due to random variation in the data
• 08:52that we can observe, but this is a mistake in error,
• 08:56and we refer to this type of error as a type one error
• 09:02or the alpha level of the test.
• 09:05We like to keep this type of error low,
• 09:08so we typically set the type one error of our tests at 5%.
• 09:13So when you're testing a single endpoint,
• 09:16you can maintain a type one error of 5%.
• 09:20However, suppose we're testing two primary endpoints
• 09:23and success on either endpoint would lead
• 09:26to a conclusion of a treatment difference.
• 09:30The type one error rate on each endpoint compounds
• 09:34and there's an inflation of the overall type one error
• 09:36probability above 5%.
• 09:40This increases the chance of false conclusions
• 09:43regarding the efficacy of the intervention.
• 09:46Special statistical testing procedures
• 09:49need to be used to control the type one error rate
• 09:52for the study with multiple endpoints.
• 09:56Multiple primary endpoints occur in three ways.
• 10:00The first is when there are multiple endpoints
• 10:03and each endpoint could be sufficient
• 10:05on its own to establish the efficacy
• 10:07of the intervention being tested.
• 10:10These multiple endpoints correspond
• 10:11to multiple chances of success,
• 10:14so failure to adjust for multiplicity
• 10:17can lead to type one error rate inflation
• 10:20and a false conclusion of effectiveness.
• 10:23The second option is when the determination of effectiveness
• 10:27depends on success on all primary endpoints
• 10:31when there are two or more primary endpoints.
• 10:34In this setting, there are no multiplicity issues related
• 10:37to the primary endpoints
• 10:40as there is only one path that leads
• 10:42to a successful outcome for the trial and therefore,
• 10:46no concern with type one error rate inflation.
• 10:50The third option combines several aspects
• 10:52of effectiveness into a single primary composite endpoint.
• 10:57This avoids multiple endpoint related multiplicity issues.
• 11:01In many cardiovascular studies
• 11:04it's common to combine several endpoints.
• 11:06For example, cardiovascular death, heart attack and stroke
• 11:11into a single composite primary endpoint.
• 11:15In this case, death is considered on its own
• 11:18as a secondary endpoint.
• 11:19If any one of the elements
• 11:21of the composite outcome is observed,
• 11:23then the endpoint has occurred for that patient.
• 11:27It's important that the endpoints included
• 11:29in the composite endpoint
• 11:31are of similar clinical importance.
• 11:34Using a composite endpoint is helpful
• 11:37when the components are individually rare
• 11:41so choosing a composite endpoint allows you to
• 11:43observe more events.
• 11:45A limitation of using a composite endpoint is that
• 11:49given the sample size of the study,
• 11:51there may not be adequate statistical power
• 11:55to test each component of the endpoint separately.
• 11:59We'll discuss statistical power in a future video
• 12:02on elements of sample size calculations.
• 12:05We'll also discuss endpoints and variables in general,
• 12:09from a data collection perspective in a future video.
• 12:15In this video, we discussed important things
• 12:18to consider when formulating your research question.
• 12:22From the research question will flow
• 12:24the specific statistical hypotheses to be tested,
• 12:28the design of the study, including the sample size,
• 12:31the data necessary to answer the research question,
• 12:35the statistical analysis that will be performed
• 12:38and the conclusions that can be made.
• 12:41The next video, which is the second video in this series,
• 12:45will give you an overview
• 12:46of study designs commonly used in clinical research.
• 12:51In video three, we will discuss the data collection process
• 12:55and formally define different variable types.
• 12:58This video will prepare us
• 13:00for video four on sample size determination.