Results: applications and quantitative synthesis

3.11 Application I: aromatherapy and essential oils

Anxiety. Donelli 2019 is the largest synthesis (65 RCTs, 7,993 people; 37 RCTs quantitative). Inhaled lavender reduces anxiety with g = −0.73 (−1.00; −0.46; 24 RCTs; I² = 85%); restricted to RCTs not at high risk of bias, g = −0.64 (−0.82; −0.47; I² = 0%) [101]. Three conditions reduce the weight of the result: (1) 58/65 RCTs (89%) are at high risk of bias when performance bias is counted — blinding odour is almost impossible; (2) the effect is significant only in high-anxiety situations (preoperative, dental; g = −0.83), not in mild anxiety; (3) for STAI-S, excluding high-risk RCTs reduces the effect from MD −5.99 to −3.67 (p = 0.05, borderline) [101].

Silexan (oral). Quite unlike the inhaled route, Silexan is a pharmacological preparation tested to drug standards. Kasper 2014 (double-blind, double-dummy RCT, n = 539): 160 mg/day reduced HAM-A by 4.0 points more than placebo (1.9–6.1; d = 0.50), 80 mg by 2.8 points (d = 0.37), whereas paroxetine 20 mg did not reach significance in the primary analysis (d = 0.21) [102]. Pooling 5 RCTs (n = 1,173): MD −2.90 HAM-A (I² = 74%) [101]. All RCTs come from the same group and were funded by the manufacturer; the mechanism is systemic absorption, unrelated to olfaction.

Expectation-controlled experiments. Kiecolt-Glaser 2008 (n = 56, within-subject crossover, randomised, expectation-controlled) is the best-designed trial in the dataset: lemon consistently increased positive affect across 4 measures (time × odor F(6,240) = 4.43) and maintained elevated norepinephrine after cold stress; lavender was no different from water on every mood measure; no odour changed cortisol, pain, wound healing or IL-6; the DTH immune response was even reduced with lavender [105]. Howard & Hughes 2008 (n = 96, double-blind): lavender did not differ from placebo in skin conductance, whereas expectation instructions had an effect — in a counter-intuitive direction [106]. Herz 2009, after screening 18 studies meeting the criteria, concluded that the mood effects of odours operate mainly through learnt associations and expectation; for example, “relaxing”/“stimulating” instructions determined the direction of heart-rate change regardless of whether the odour was real or placebo (Campenni 2004, n = 90) [98].

Sleep. Lin 2019 (31 RCTs) reported g = 1.10 (0.81–1.39) — but I² = 91%, publication bias was significant (Egger p = 0.004; Begg p = 0.008) and only 4/31 RCTs were at low risk of bias [104]. A large effect amid such heterogeneity and bias should be understood as a signal, not an estimate. Goel 2005 (PSG, n = 31) found that lavender increased the proportion of SWS [108]; Perl 2016 found that all odours — including unpleasant ones — increased delta during NREM [109], suggesting a non-specific effect for lavender.

Dementia and depression. A Cochrane review (13 RCTs, n = 708) rated the evidence for agitation as very low: of the 5 more reliable trials, 4 found no effect; Burns 2011 (same lemon balm oil, better design) did not replicate the positive result of Ballard 2002 [100]. For depression (12 RCTs, n = 1,226), the highest-quality trials (Jadad 4–5) were all negative; positive results were concentrated in Jadad 1–2 [103].

3.12 Application II: olfactory training

Hummel 2009 (n = 56, non-randomised): 28% of the group training with 4 odours × 2 times/day × 12 weeks achieved a clinically significant TDI improvement vs 6% of the non-training group [110, 112]. Sorokowska 2017 meta-analysis (13 studies): TDI g = 1.10 (0.46–1.73); identification g = 0.83, discrimination 0.89, but threshold only 0.34 (0.10–0.57); post-infectious smell loss benefited most (g = 1.52); training duration predicted improvement (β = 0.57); healthy older people did not improve [111]. Kattar 2021 (k = 4): OR = 2.77 (1.67–4.58) for achieving clinically significant improvement [113]. Threshold improvement was weaker in patients with olfactory dysfunction (g = 0.15) than in other groups (g = 0.64) [111]. The pattern “identification/discrimination improve a lot, threshold improves little” suggests (an inference by the authors of this review) that most of the benefit comes from central learning (naming, attention, categorisation) rather than peripheral regeneration. Time-course data: 1.7% after 1 month → 26.7% after 3 months → 41.7% after 6 months; everyday odours are as effective as the standard kit; high concentrations are better than low [112]. Spontaneous recovery is only ~6–8% in ~4 months [110] — but this estimate comes from a very small control group (n = 16).

3.13 Application III: scent marketing and brand memory

Roschk & Hosseinpour 2020 (71 samples, 15,447 people, 671 effects) provide the most reliable estimate: ambient scent increases recall the most (r = 0.177; +11.3%), increases evaluations and mood less, and does not significantly increase spending (r = 0.051; k = 52) [117]. Evaluation + memory are larger than mood (p = 0.023) — the authors conclude that scent in retail is a cognitive rather than an emotional stimulus. Moderators: scent congruent with the context (b = 0.065), simple scents (multidimensional scents weaker, b = 0.079), incongruent music reduces the effect; the effect saturates when 60–80% of customers detect the scent [117]. Pleasantness has no main effect — consistent with Spangenberg 1996: neutral and pleasant scents act equally, not via mood; scented stores make perceived time shorter (576 vs 661 s) while actual time is unchanged [116].

Krishna 2010 showed an entirely different mechanism: scent attached to the product increased recall of product attributes and was retained after 2 weeks (77–92% retained vs 27% unscented; η² = 0.36), fully mediated by the number of times the scent was mentioned at retrieval, not via mood; ambient scent had no effect [118]. This is the applied version of the principle of odour–object associative encoding [87, 88]. Weber & Heuberger 2008: in 4 field experiments (n = 32), plant scents improved calmness, alertness and mood but depended on the order of conditions; in the artificial-odour experiment (n = 22), pleasant odours did not change mood whereas unpleasant odours worsened it — a valence asymmetry [119].

3.14 Quantitative synthesis

Figure 8 places 19 estimates on the same standardised scale. Three clusters emerge: (i) small, precise cognitive–memory effects (TMR, chemosignals, retail recall: d ≈ 0.1–0.4, narrow CIs); (ii) large but uncertain clinical effects (sleep, anxiety, olfactory training: g ≈ 0.6–1.1, wide CIs, high heterogeneity — I² 85–91% for sleep and anxiety, Q = 172.7 for olfactory training); (iii) null or negative effects (TMR during REM, retail spending, odour spatial memory compared with vision). The classic rule “the larger the effect, the smaller and less controlled the studies” is clearly visible: estimates corrected for bias (TMR 3PSM) or restricted to low-risk studies (lavender) are always smaller than the crude estimates.

Scientific reference material, not medical advice. Figures are redrawn from published data, not copied from the original figures. GRADE and risk-of-bias ratings are the compilers’ own. Every number traces back to detailed reading notes on each paper.