From Raw Papers to Research Protocol: A Step-by-Step Guide
The distance between a stack of relevant papers and a finished protocol is longer than it looks. This is about what happens in that space.
The distance between a stack of relevant papers and a finished research protocol is longer than it looks. Most researchers have experienced the specific frustration of knowing the literature well, having read the key studies, having a clear sense of the gap, having a hypothesis that feels right, and then sitting down to write the protocol and finding that the path from what you know to what you need to write is not as direct as you expected.
This guide is about that distance. Not about what a protocol should contain, there are plenty of templates for that, but about the process of building one from the ground up, informed by the evidence, in a way that holds up to scrutiny.
Step one: define the question before you search
This sounds obvious. It is often skipped. The temptation is to start searching the literature immediately, on the reasonable grounds that you need to know what is out there before you can define what you are asking. The problem with this approach is that an undefined question produces an undirected search, and an undirected search produces an overwhelming result that is difficult to synthesise.
Start with the clinical or scientific uncertainty you are trying to resolve. Write it in plain language, without worrying about PICO structure yet: "We do not know whether X intervention is effective in Y population because the existing studies have all been done in Z population, which differs in the following ways." That sentence, or something like it, is the foundation of your protocol. Everything that follows should be traceable back to it.
Then structure it into PICO or the appropriate framework for your question type. The structure is not the question, it is a tool for making the question searchable and comparable to existing evidence.
Step two: map the evidence landscape, not just the evidence
There is a difference between a literature search and a literature map. A search returns papers. A map tells you where the evidence is strong, where it is weak, where it is contested, and where it is absent. A protocol built on a search may miss the structural features of the evidence base that determine what kind of study design is appropriate and what outcomes are worth measuring.
Mapping the evidence means looking at the literature at multiple levels. At the level of individual studies: what populations have been studied, what interventions tested, what outcomes measured, what methodological approaches used. At the level of existing syntheses: what systematic reviews exist, how current they are, what their conclusions were, what gaps they identified. At the level of ongoing research: what trials are currently registered in ClinicalTrials.gov or similar registries, what questions are already being studied, what results are expected and when.
This last point is underappreciated. Designing and running a study that is answered by an ongoing trial's results, results that will be published before yours, is an avoidable waste of resources. Trial registry searches should be a standard part of protocol development, not an afterthought.
Step three: let the methodology follow the question
One of the most common protocol weaknesses we see, and we review a lot of protocols at NousLab, is the mismatch between the question and the design. The question asks about causal efficacy; the design is observational. The question is about rare outcomes; the sample size is calculated for a common one. The question involves a heterogeneous population; the eligibility criteria select a homogeneous subset that may not represent it.
The methodology section of a protocol is an argument that a particular design can answer a particular question. It only works if the design is chosen to fit the question, not the other way around. The pragmatic constraints, available sample size, budget, follow-up period, are real and should be addressed honestly. But they should be addressed as constraints on what is achievable, not as drivers of what is studied.
If the constraints mean you cannot answer the question you set out to answer, that is a finding. It means either that the question needs to be reformulated as something that is answerable within your constraints, or that the study should not be done yet. Both outcomes are better than running a study that produces uninterpretable results.
Step four: choose outcomes that the evidence base can support
Outcome selection is where the literature review most directly informs the protocol design. The outcomes you choose need to be measurable, clinically meaningful, and sufficiently sensitive to detect the effect you expect, if one exists.
The last of these is the hardest. Effect size estimates for sample size calculations should come from the existing evidence, not from what would make the study feasible. If the published literature suggests a small effect, powering your study for a large one because you cannot afford to recruit the required sample for a small one does not make the study valid. It makes it underpowered for the actual question.
Core Outcome Sets, agreed-upon standardised outcomes for specific clinical areas, are an underused resource here. The COMET Initiative maintains a database of Core Outcome Sets that can help ensure your outcome selection aligns with what other researchers in the field are measuring, which improves comparability and reduces the outcome reporting variability that plagues meta-analyses.
Step five: write the protocol as if someone else will run the study
The practical test of a protocol is whether someone who was not involved in writing it could run the study from the document alone and produce results that are comparable to what you would have produced. Most protocols fail this test, not because of conceptual problems, but because of specificity gaps, procedures described at a level of generality that leaves too much to interpretation.
Eligibility criteria are the most common specificity problem. "Patients with significant renal impairment" is not an eligibility criterion. "Adults aged 18 or over with eGFR less than 45 ml/min/1.73m² on two measurements at least 90 days apart" is. The specificity is not bureaucratic pedantry, it is what makes the study reproducible and the results interpretable.
This is one of the areas where AI-assisted drafting is genuinely useful. NousLab can generate eligibility criterion drafts informed by the specificity standards used in published trials in the same area, giving researchers a concrete starting point that reflects current methodological practice rather than a template that requires heavy adaptation. The researcher still makes every substantive decision. The scaffolding is already there.
The protocol as a living document
A protocol is not finished when it is submitted to an ethics committee. It is finished when the study is finished, and even then it leaves a trace in the published methods section that will be scrutinised by every reader and reviewer who engages with your results.
The effort invested in grounding a protocol in the evidence base, in mapping the literature carefully, choosing the design deliberately, specifying the outcomes precisely, is effort that pays forward through every subsequent stage of the research process. Studies that start with a well-built protocol have fewer amendment requests, cleaner data, and more interpretable results.
That is not a coincidence. A good protocol is evidence that someone thought carefully before they started. That kind of thinking does not stop mattering once the study is underway. See how NousLab supports researchers at the protocol development stage.
