Design the test

01One change at a time

If two things change together, a difference can’t be pinned on either, and that goes for changes outside the plan too, such as a new food that day. One self-experiment study held everything else steady: people ate the same breakfast on every test day, adding or leaving out only the food being tested (Karkar et al., 2017).

A whole food is several things at once: foods that contain gluten often contain fructans too, a kind of carbohydrate, so a test of such a food tests both (Skodje et al., 2018).

02Repeat it

One stretch before a change and one after it is what single-case research calls an AB design. With nothing repeated, other explanations are hard to rule out, so its standards ask for at least three chances to show an effect, at three different times, each with at least three readings (Kratochwill et al., 2010). The panel calls three a professional convention, not a figure with a formal basis.

A US guide to one-person trials puts it simply: one switch gives almost no protection against chance, so the sequence has to be repeated (Kravitz and Duan, 2014). Repeats also show whether a difference is steady. In a simulated example, one comparison per person looked the same whether some people responded reliably and others mostly didn’t, or everyone responded only some of the time; a second comparison told the two apart (Senn, 2016). In a review of 74 published randomised one-person trials, the median number of periods was six (Hawksworth et al., 2024).

03Let chance, or a balanced plan, set the order

Test days you choose can line up with something else: weekends, a busy week, days that are already going well. The US guide warns that days chosen this way can bias the comparison (Kravitz and Duan, 2014).

A 1986 paper set out the classic answer: pairs of periods, one on each option, in an order chosen at random (Guyatt et al., 1986). If the two options are A and B, a run like AAAABBBB gives no protection against a slow drift over the weeks; ABBAABBA does better, and ABBABAAB guards against both steady and less regular drifts (Kravitz and Duan, 2014).

Current key points for these trials recommend shuffling within pairs, so a run like AAAABBBB can’t happen, though the order becomes easier to guess (Chatters et al., 2024). Pure chance has its own risk: in the self-experiment study, a fully random order sometimes gave three or four days of the same kind in a row, or bunched one kind on weekdays or weekends (Karkar et al., 2017). The US guide calls a balanced order the more reliable choice, and a fully random one harder to guess (Kravitz and Duan, 2014).

04Gaps between periods, and how long each lasts

Some effects linger into the next period and blur the comparison: this is carryover. Key points for one-person trials say it should be gone before the next period is judged, with a gap, called a washout, where it is likely (Chatters et al., 2024). Carryover can also be memory, as people recall how they felt in earlier periods (Chatters et al., 2024).

One food study set its gaps by watching symptoms: at least seven days, and longer if symptoms from the previous week had not settled (Skodje et al., 2018).

Each period must be long enough for an effect to show (Chatters et al., 2024). The US guide’s example used 7-day periods, partly because shorter ones could be skewed by differences between weekdays and weekends (Kravitz and Duan, 2014). When it isn’t known how long an effect lasts, it suggests longer periods with daily ratings. In the self-experiment study, which treated each day as a fresh start, some people felt effects build up over as long as three days (Karkar et al., 2017).

05What the studies found

• A 1986 paper set out a way to test within one person: repeated pairs of periods, one on the active option and one on a placebo or an alternative, run double-blind in a random order (Guyatt et al., 1986).
• A 2014 user’s guide to one-person (n-of-1) trials from the US Agency for Healthcare Research and Quality names three main design elements: a balanced order, blinding and systematic measurement (Kravitz and Duan, 2014). It says one switch from A to B gives almost no protection against chance, and that the method suits options that start working and wear off fairly quickly.
• In 2010, a panel set standards for single-case studies for the US What Works Clearinghouse (Kratochwill et al., 2010). A study must try at least three times, at different points in time, to show an effect, with at least three readings per phase, so a plain AB design does not meet the standard.
• A 2016 statistics paper argues that telling a steady personal response apart from variation between occasions needs comparisons repeated in the same person (Senn, 2016).
• A 2024 review of 74 randomised one-person trials published from 2011 to 2023 found a median of six periods, each lasting a median of 14 days (Hawksworth et al., 2024). Fewer than half (43%) used a washout, with a median gap of 7 days.
• A 2024 set of 22 key points for one-person trials, drawn up with clinicians, researchers and patient representatives, recommends shuffling the order in small blocks, blinding where feasible, and a washout where carryover is likely (Chatters et al., 2024).
• In a 2017 study, 15 adults with irritable bowel syndrome each ran a 12-day test of one food, using a phone app (Karkar et al., 2017). The app chose 6 days at random to eat the food at breakfast and 6 to leave it out, and told people each morning which kind of day it was. The authors note that 12 days may be too few to pick up small effects.
• In a 2018 double-blind crossover study, 59 people ate muesli bars containing gluten, fructans or neither for a week each, with a break of at least seven days between, until symptoms had settled (Skodje et al., 2018).

06Sources on this page

Each source is listed once, with links to its record. The reference list says what was read for each, and when it was checked.

1. Chatters et al., 2024Chatters R, Hawksworth O, Julious S, Cook A. 2024. The development of a set of key points to aid clinicians and researchers in designing and conducting n-of-1 trials. Trials 25(1):473.DOI 10.1186/s13063-024-08261-z · PubMed 38992786 · PMC11241860
2. Guyatt et al., 1986Guyatt G, Sackett D, Taylor DW, Chong J, Roberts R, Pugsley S. 1986. Determining optimal therapy—randomized trials in individual patients. New England Journal of Medicine 314(14):889–892.DOI 10.1056/NEJM198604033141406 · PubMed 2936958
3. Hawksworth et al., 2024Hawksworth O, Chatters R, Julious S, Cook A, Biggs K, Solaiman K, et al. 2024. A methodological review of randomised n-of-1 trials. Trials 25(1):263.DOI 10.1186/s13063-024-08100-1 · PubMed 38622638 · PMC11020886
4. Karkar et al., 2017Karkar R, Schroeder J, Epstein DA, Pina LR, Scofield J, Fogarty J, et al. 2017. TummyTrials: A Feasibility Study of Using Self-Experimentation to Detect Individualized Food Triggers. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI), 6850–6863.DOI 10.1145/3025453.3025480 · PubMed 28516175 · PMC5432136
5. Kratochwill et al., 2010Kratochwill TR, Hitchcock J, Horner RH, Levin JR, Odom SL, Rindskopf DM, et al. 2010. Single-case designs technical documentation. What Works Clearinghouse, US Institute of Education Sciences. Version 1.0 (pilot), June 2010.ies.ed.gov
6. Kravitz and Duan, 2014Kravitz RL, Duan N, editors; DEcIDE Methods Center N-of-1 Guidance Panel. 2014. Design and Implementation of N-of-1 Trials: A User's Guide. Agency for Healthcare Research and Quality. AHRQ Publication No. 13(14)-EHC122-EF. Rockville, MD.effectivehealthcare.ahrq.gov
7. Senn, 2016Senn S. 2016. Mastering variation: variance components and personalised medicine. Statistics in Medicine 35(7):966–977.DOI 10.1002/sim.6739 · PubMed 26415869 · PMC5054923
8. Skodje et al., 2018Skodje GI, Sarna VK, Minelle IH, Rolfsen KL, Muir JG, Gibson PR, et al. 2018. Fructan, Rather Than Gluten, Induces Symptoms in Patients With Self-Reported Non-Celiac Gluten Sensitivity. Gastroenterology 154(3):529–539.e2.DOI 10.1053/j.gastro.2017.10.040 · PubMed 29102613