Why pre-register

A study whose analysis is decided after the data are seen can find almost anything. Pre-registration fixes the question, the outcomes, the sample size, the analysis and the rules for reading the result before any participant is enrolled, and publishes them. This chapter is that document for the question this book keeps returning to: in healthy adults, does a 16:8 eating schedule change sustained attention? It follows the structure of the SPIRIT guidance for trial protocols (Chan et al. 2013) and will be reported according to the CONSORT extension for crossover trials (Dwan et al. 2019).

Design

A randomised, two-period crossover trial. Each participant spends four weeks on each of two schedules, in random order, separated by a two-week washout of habitual eating.

The 11:00–19:00 window is chosen so that the late-morning test falls near the end of the 16-hour fast, and so that the evening meal and sleep times are similar in both periods.

Participants

Inclusion. Adults aged 25–60; body-mass index 18.5–30 kg/m²; regular sleep of at least 6 hours a night; habitual eating window of 12 hours or more; able to use the study application.

Exclusion. Diabetes of any type, or use of any medicine that lowers blood sugar; use of an SGLT2 inhibitor; pregnancy, planned pregnancy or breastfeeding; current or past eating disorder, or a positive screen on a validated eating-disorder questionnaire; kidney or liver disease, gout, heart disease, or a history of fainting; shift work or travel across more than two time zones during the study; current regular practice of time-restricted eating or fasting; regular use of ketone or NAD⁺-precursor supplements; any condition for which a study physician judges fasting unsafe.

Outcomes

Primary outcome. Lapses (responses slower than 355 ms) on the 3-minute psychomotor vigilance test (Basner et al. 2011), taken on the study application at 10:30 each day, averaged over the final 7 days of each period.

Secondary outcomes.

1. Capillary β-hydroxybutyrate at 10:30 on the final 7 days of each period (manipulation check).
2. Mean reaction time on the same test.
3. Hunger (0–10) and positive and negative mood (short validated scale) at 10:30.
4. Sleep duration from a wrist-worn activity tracker.
5. Adverse events, including every warning sign defined in Chapter 8.

Hypotheses

H1 (primary). 16:8 eating reduces late-morning lapses compared with habitual eating, by at least the smallest effect of interest.

H2 (manipulation check). Late-morning β-hydroxybutyrate is higher during 16:8 than during habitual eating.

H3 (mechanism, secondary). Within the 16:8 period, a participant’s daily β-hydroxybutyrate predicts fewer lapses that day, after accounting for the previous night’s sleep. (Chapter 3 predicts that it will not.)

H4 (harm, secondary). Hunger and negative mood at 10:30 are higher during 16:8.

Sample size

The smallest effect of interest is a within-person standardised effect of $d_z = 0.30$: smaller effects on a 3-minute test would not justify changing a daily routine. With two-sided α = 0.05 and 90 per cent power:

import random
from statistics import NormalDist, mean, stdev

z = NormalDist().inv_cdf
alpha, power = 0.05, 0.90
d_z = 0.30      # smallest effect of interest, within-person standardised
# normal-approximation formula plus a small correction for the t distribution
n = ((z(1 - alpha / 2) + z(power)) / d_z) ** 2 + z(1 - alpha / 2) ** 2 / 2
print(f"completers needed: {n:.1f} -> {int(n) + 1}")
print(f"to enrol, allowing 20% drop-out: {round((int(n) + 1) / 0.8)}")

# Check by simulation: 2,000 crossover trials with 120 completers and a true effect of d_z = 0.30.
rng = random.Random(9)
hits = 0
for _ in range(2000):
    diffs = [rng.gauss(d_z, 1) for _ in range(120)]
    se = stdev(diffs) / len(diffs) ** 0.5
    hits += abs(mean(diffs) / se) > 1.980        # two-sided t test, 119 degrees of freedom
print(f"simulated power with 120 completers: {hits / 2000:.3f}")
# completers needed: 118.7 -> 119
# to enrol, allowing 20% drop-out: 149
# simulated power with 120 completers: 0.898

The trial will enrol 150 participants to finish with at least 120.

Analysis plan

The primary analysis uses all randomised participants with data from both periods. The mean number of lapses in the final 7 days of each period is analysed with a linear mixed model with schedule, period and order as fixed effects and participant as a random effect; the estimate of interest is the difference between habitual eating and 16:8 (positive = fewer lapses on 16:8), standardised by the within-person standard deviation of the differences, with a 95 per cent confidence interval. Carry-over is guarded against by the washout and by using only the final week of each period; the order effect is reported. A sensitivity analysis uses all participants under multiple imputation. H3 uses a multilevel model of daily data within the 16:8 period. No other analyses will be described as confirmatory.

The decision rule, written as code

The result will be read by the rule below, which is fixed now.

SESOI = 0.30          # smallest within-person effect worth having, in standard deviations

def verdict(lo, hi):
    """lo, hi: 95% confidence interval of (lapses on habitual eating - lapses on 16:8), standardised.
    Positive = fewer lapses on 16:8."""
    if lo > 0:
        return "benefit supported"
    if hi < 0:
        return "harm: attention worse on 16:8"
    if hi < SESOI:
        return "benefit refuted: an effect worth having is ruled out"
    return "inconclusive: the study could not tell"

examples = {"result A": (0.08, 0.44), "result B": (-0.12, 0.25), "result C": (-0.20, 0.41),
            "result D": (-0.48, -0.07)}
for name, (lo, hi) in examples.items():
    print(f"{name}  CI [{lo:+.2f}, {hi:+.2f}] -> {verdict(lo, hi)}")
# result A  CI [+0.08, +0.44] -> benefit supported
# result B  CI [-0.12, +0.25] -> benefit refuted: an effect worth having is ruled out
# result C  CI [-0.20, +0.41] -> inconclusive: the study could not tell
# result D  CI [-0.48, -0.07] -> harm: attention worse on 16:8

The examples are invented to show how each kind of result will be read:

• The popular claim is supported only if the whole interval lies above zero.
• The claim is refuted if the interval excludes any benefit as large as 0.30: a 16:8 schedule may still be useful for other reasons, but not because it sharpens late-morning attention.
• A harm is reported as a harm if the interval lies wholly below zero.
• An inconclusive result is reported as inconclusive, not as “a trend”.

The book’s own prediction (Chapter 6) is result B: no effect worth having.

Safety

Participants receive the stop rules of Chapter 8 in writing and in the application. A study physician reviews all adverse events weekly. A participant who has two warning days in one week, or any fainting, stops the 16:8 schedule and is followed up; their data are analysed as randomised. The trial will be stopped by an independent safety committee if serious adverse events are more frequent during 16:8 than during habitual eating.

Ethics, registration and conflicts of interest

The trial will not enrol anyone before approval by an independent research ethics committee and registration on a public clinical trial registry; the approval number and registration identifier will be added to the next version of this book. It follows the Declaration of Helsinki (World Medical Association 2025). Each participant gives written informed consent and can withdraw at any time without giving a reason. Participants are paid for their time, not for completing the fasting schedule.

What this trial cannot show

It cannot show whether fasting affects long-term brain health, autophagy or ageing; four weeks and a reaction-time test are not designed for those questions. It tests one schedule, in healthy adults who volunteer, many of whom will already be interested in fasting. A null result would not show that 16:8 is useless; it would show that it does not sharpen late-morning attention by an amount worth having. Whatever the result, it will be published, and the next version of this book will be revised to match it.