Opens the treatment question for the sleep cluster: once someone has chronic insomnia, what should
they do? Three interventions, three evidence types. The beyond-summary move is that the ranking is not
set by effect size on a sleep number — it is set by evidence type, durability, and harm, which run in
the opposite direction to how these are usually reached for (a drug first, behaviour last).
(Ferracioli-Oda et al., 2013; Kripke et al., 2012; inferred from van Straten et al., 2018)
Why the three cannot be ranked by effect size — the commensurability table
Each source reports a different quantity on a different design, so a numeric league table would compare a standardized-mean-difference against minutes against a hazard ratio. The honest artifact is: decide on evidence-type + harm, not on the effect number.
| Parameter | CBT-I (van Straten) | Hypnotics (Kripke) | Melatonin (Ferracioli-Oda) | Same quantity? |
|---|---|---|---|---|
| Design | 87 RCTs, MA (gold) | 1 matched cohort (moderate) | 19 RCTs, MA (gold) | NO — RCT-MA vs observational vs RCT-MA |
| Primary effect | ISI g=0.98; SOL g=0.57; TST g=0.16 | mortality HR 3.6-5.3 (dose); cancer HR 1.35 | SOL -7.06 min; TST +8.25 min; quality SMD 0.22 | NO — g vs HR(harm) vs minutes |
| What is measured | sleep continuity/quality (benefit) | death + cancer (harm) | latency/duration (benefit, small) | NO — different endpoints entirely |
| Durability / tolerance | robust incl. chronic hypnotic users; long-term asserted | short-term efficacy only | no tolerance (effect grows with duration) | comparable direction, not magnitude |
| Harm signal | none reported | strong (confounded) | benign side-effect profile | — |
Because every same quantity? cell is NO, no cross-source magnitude claim is made. The within-source comparisons stand: van Straten’s smallest effect is on total sleep time (g=0.16) — CBT-I consolidates sleep, it does not add hours; Ferracioli’s own benchmark puts melatonin’s <7-min latency gain below benzodiazepines’ 10.0-19.6 min. (Ferracioli-Oda et al., 2013)
CBT-I — the first-line, and the only one that restores sleep without a drug [EXTRACTED]
van Straten pooled 87 RCTs (3724 treated): large effects on insomnia severity (ISI g=0.98), sleep efficiency (g=0.71), wake-after-onset (g=0.63) and onset latency (g=0.57); the smallest on total sleep time (g=0.16). (van Straten et al., 2018) Two decision- relevant properties:
- It works where drugs are the problem. Effects were «quite robust (similar for patients with or without comorbid disease, younger or older patients, using or not using sleep medication)» — including chronic hypnotic users, i.e. it is the exit route from pills.
- It is guideline first-line. «CBTI is at least as effective as pharmacotherapy» and «the American College of Physicians recently recommended CBTI as the initial treatment for all adults with insomnia». (van Straten et al., 2018)
- Cost + honesty. Face-to-face / >=4-session beats self-help; publication bias is present, so «the actual effects of CBTI on insomnia might be smaller although still significant». Adherence + access (therapist availability) are the real constraints -> Sleep Duration and Mortality.
Hypnotics — real sedation, but a harm signal that inverts the naive choice [EXTRACTED]
Kripke’s matched cohort found a strong dose-response mortality association: HR 3.60 / 4.43 / 5.32 across ascending dose tiers, and HR 3.60 even at <18 pills/year — plus a cancer signal (upper-third HR 1.35). (Kripke et al., 2012) The association «robust within groups suffering each comorbidity, indicating that the death and cancer hazards … were not attributable to pre-existing disease».
The caveat is decisive and must be stated at point of use (symmetric standards). This is a single observational cohort; the live confounder is confounding by indication — the same distress, psychiatric load and occult illness that lead to a hypnotic prescription independently raise mortality, and «residual confounding could not be fully excluded». Stratifying on 116 comorbidity combinations attenuates but cannot remove it, and no chronic-hypnotic mortality RCT exists (deemed unethical). The lead author also runs an anti-hypnotic advocacy site — reason to hold the magnitude loosely. So the hazard is a decision-relevant harm signal, not a demonstrated causal death toll: it shifts hypnotics from a first reach to a short-course, lowest-effective-dose, last-resort option. Kripke’s own conclusion: «A consensus is developing that cognitive-behavioural therapy of chronic insomnia may be more successful than hypnotics.»
Melatonin — modest, benign, no tolerance [EXTRACTED]
Ferracioli-Oda’s 19-RCT MA: melatonin cut sleep-onset latency by 7.06 min (4.37-9.75), raised total sleep time by 8.25 min (subjective; NS on objective measures), and improved quality (SMD 0.22). (Ferracioli-Oda et al., 2013) Two features earn it a place despite the small effect: it shows no tolerance (effect grew with trial duration and dose, unlike benzodiazepines), and a benign side-effect profile — «the absolute benefit of melatonin compared to placebo is smaller than other pharmacological treatments … melatonin may have a role … given its relatively benign side-effect profile». A reasonable low-stakes adjunct (especially for circadian/timing complaints -> Sleep Regularity and Mortality), never a substitute for CBT-I.
Decision relevance — the substitution frame
- Order: CBT-I first; melatonin as a benign adjunct; hypnotics short-course, last. The usual order is inverted because the ranking follows harm + durability, not the effect number: CBT-I carries the large, durable, harm-free benefit and is the exit from pills; melatonin is small but safe and non-tolerance-forming; hypnotics buy short-term sedation against a real (if confounded) mortality/cancer signal.
- Do sleep aids give real sleep? CBT-I restores continuity and efficiency (the sleep-architecture answer); melatonin nudges timing/latency; sedative hypnotics produce drugged sedation whose long-run safety is unestablished. The three are not interchangeable routes to one outcome.
- Stratum. Chronic insomnia is the target; the chronic hypnotic user is the highest-value stratum for a CBT-I switch (the one group where the harm lever and the benefit lever both point the same way).
Limits
- No head-to-head trials among the three — the ordering rests on separate evidence bases + a harm asymmetry, not a direct comparison; the durability-of-CBT-I-vs-hypnotics claim is asserted from component reviews, not pooled here.
- The harm anchor is one confounded cohort. The hypnotic-mortality magnitude is the weakest-design input; it is used as a labelled counterweight (moderate tier, gold-gate override), not a precision figure -> Sleep Duration and Mortality.
- All benefit endpoints are surrogate/self-report (ISI, PSQI, latency) — no hard-outcome trial shows treating insomnia lowers mortality. Coherence, not validity (R1).
- AWAITS the ACP/AASM insomnia guideline as a second guidance family, and any hypnotic-vs-placebo long-term safety RCT (none exists).
(Ferracioli-Oda et al., 2013; Kripke et al., 2012; inferred from van Straten et al., 2018)