Field notes on frequently asked study design questions that don’t have easy answers.

In our conversations with researchers either starting a new study or planning their first study, we’ve noticed an interesting trend before the study starts and after the study ends. The questions investigators ask before a study begins are often more important than the answers they eventually publish.

Unfortunately, many investigators soon discover that biospecimens have shown little interest in traveling backward through time. As assay technologies continue to evolve, new biomarkers emerge, and novel mechanistic questions often arise long after a study begins. In fact, the samples collected today may ultimately support analyses that were not even conceivable when the protocol was written.

In our line of work, we talk to the future-you often, and we know the future-you can be an exceptionally demanding collaborator. A single additional timepoint, a slightly different sampling schedule, or one more aliquot can become surprisingly valuable once the data begin revealing new questions. Whether a study ultimately supports a single publication or continues generating new discoveries for years, its long-term potential is often determined months before the first participant ever walks through the door.

1. High-value cohorts are scientific platforms

The rapid growth of biomarker research has made it easier than ever to measure biology. It has also made it easier to confuse two very different scientific goals: explaining why something happens and predicting what will happen next.

Mechanistic studies seek to understand biological pathways and causal relationships. Predictive studies focus on identifying patterns that reliably forecast an outcome, even when the underlying biology remains incomplete. Both approaches are valuable, but they often require different study designs, sampling strategies, and analytical decisions.

Before future-you inherits this study protocol, it is worth asking the present you a simple question: what is this study ultimately trying to accomplish?

• Is the goal to discover novel biology?
• To test a mechanistic hypothesis?
• To develop a predictive biomarker?
• Or to evaluate an intervention?

The answer influences everything from cohort selection and sampling schedules to specimen handling, metadata collection, and statistical planning. A study optimized for one objective may be poorly suited for another.

2. Specimen strategy often outweighs assay capability

For many biomarkers, the question is no longer what can be measured reliably. The more consequential question is whether the samples were collected and preserved correctly in the first place.

The question you want to ask here is how you are negotiating the tradeoff between deeper phenotyping and broader recruitment. Every additional specimen, timepoint, or metadata field comes with a cost. The challenge is deciding where additional information meaningfully increases scientific value and where it simply increases complexity.

3. Multi-matrix collection enables broader questions

The choice of biological matrix should follow the biological question, not the other way around. Some questions are well addressed by saliva alone. Others require information from multiple physiological systems. As research increasingly examines interactions between endocrine, immune, metabolic, and cardiovascular pathways, combining saliva with complementary specimen types such as dried blood spots can provide a more complete picture of human biology. The goal is not simply to expand the list of analytes. It is to design studies that better reflect the interconnected nature of human biology.

4. Designing for scientific yield

Recruiting participants is often the most difficult, time-consuming, and expensive part of a study. Every participant represents an opportunity to learn more than a single primary endpoint.

Budget constraints are real, but maximizing scientific value does not always require collecting more samples. Sometimes it means collecting them more thoughtfully. Participant-centered collection protocols can improve compliance and specimen quality. A well-designed power analysis can help avoid studies that are too small to answer the question being asked, or unnecessarily large for the effect being measured.

Creative study design can also extend the value of existing specimens. Depending on the research question, investigators may choose to analyze selected biomarkers in only a subset of participants, evaluate exploratory biomarkers on representative cohorts, or take advantage of high-precision laboratories that can justify singlet testing under appropriate circumstances. Not every biospecimen needs to answer every question, but every biospecimen should have the opportunity to answer an important one.

5. Sample beyond statistical significance

Statistical significance answers whether an observed difference is likely to have occurred by chance. It does not necessarily answer whether that difference is meaningful for an individual participant.

If the study aims to evaluate intervention effects or longitudinal change, it is worth deciding in advance what constitutes meaningful change. The Reliable Change Index (RCI) provides one framework by accounting for analytical precision and biological variability when interpreting individual-level change over time.

Thinking about meaningful change during study design should be core to your assay selection, sampling schedules, power calculations, and follow-up intervals. Like many aspects of study design, these decisions are much easier to make before the first specimen is collected than after the data have been analyzed.

6. Power analyses need saliva-specific assumptions

Determining a sample size always comes down to a mix of research design, budgetary, and practical constraints. Every one of us has weighed these tradeoffs many times, in training and in day to day work.

What is less obvious is what the published literature actually shows. A look back at biobehavioral research suggests that most effect sizes are small to modest, and that published studies in this space are more often than not underpowered to detect them.

The reasons vary from study to study, but the consequence is fairly consistent. The relationships we report may only be the tip of the iceberg. If more studies were adequately powered, there is no telling what else we would find. This problem becomes even more pronounced once a study is broken into subgroups: older versus younger, male versus female, responders versus non-responders, low versus high whatever. Power within each of those smaller cells drops even further. Most of us also have a file drawer, or a folder, of studies that did not reach statistical significance and were never published.

Every study plans its sample size around a power analysis, and every power analysis rests on assumptions. Based on our experience in salivary bioscience, here is what we think is worth knowing before you finalize yours:

• Error variance typical of salivary bioscience data
• How participant burden during saliva collection drives missing data
• Drop-out risk that can leave you short of the endline sample size you proposed at the outset
• What effect size actually looks like in a salivary bioscience application
• Using the Reliable Change Index inside the power analysis itself, not only after the data are in
• The ratio of responders to non-responders, and what that means for your capacity to study recovery

None of these live inside a standard power calculator, but they belong in yours. If you want to talk through what that looks like for your study, we’re happy to have that conversation with you.

7. What a well-powered study actually costs

Everyone knows a power analysis starts with an effect size. Fewer people know how small that effect size actually is once you leave the lab and start measuring what happens after someone wakes up.

Take the Cortisol Awakening Response, or CAR, the rise in cortisol during the first 30 to 45 minutes after waking. Published research puts the typical group difference for something like depression at a modest r of .15. Run the numbers and that means 172 people per group, 344 total, just to detect it at 80% power. Most researchers stop there. They shouldn’t.

CAR has to be collected at home, on a schedule, by the participant. The consensus guidelines for CAR report real compliance data: waking samples come in more than 15 minutes late about 7% of the time, and the second sample misses its window about 15% of the time. Put those together and a single collection day has roughly an 80% chance of being usable. That is before anyone drops out of the study.

The guidelines call for at least 2 valid days per person to get a reliable estimate. Ask participants for exactly 2 days and you are betting on both of them landing clean. Ask for 3 days instead, and the odds of getting 2 valid ones jump to about 89%. That single extra day of data collection is one of the cheapest things you can add to a protocol.

Now stack on attrition. A typical multi-day, at-home protocol loses about 20% of participants before the study ends. Once you account for attrition and the chance of falling short of 2 valid days, that original target of 344 analyzable participants turns into an enrollment target closer to 480, about 240 per group. Skip the 3-day buffer and ask for only 1 day of collection, and the number climbs even higher.

None of this shows up in a standard power calculator. It shows up in your consent forms, your recruitment budget, and your timeline. The good news is that the fix is usually small. A well-placed extra collection day is far cheaper than recruiting dozens of additional participants to make up for the ones a single missed sample quietly cost you.

The Bottom Line

Unfortunately, the future-you will most likely critique the present-you’s protocol, but just like your biospecimens, won’t travel back in time to improve it. Every protocol is, in some small way, a prediction about the future. If working with researchers has taught us anything, it’s that today’s complete dataset often becomes tomorrow’s “if only we’d collected…”

If you’re weighing these considerations in an upcoming study, we’re always happy to discuss study design, sampling strategies, biomarker selection, or protocol development. Sometimes a short conversation early in the planning process can open opportunities that are difficult, or even impossible, to recover later. Reach out and we’ll help you talk it through.