Facet of the chronic-pain cluster (nucleus: Chronic Pain and Physical Activity). The nucleus asks
whether to exercise at all; this page asks a prescription question inside that choice — must the
exercise stay pain-free? Scope: a modifiable function/pain lever in adults with localized chronic
musculoskeletal pain. Out of scope: diagnosing the condition, injection or surgery choice.
The evidence — one meta-analysis, seven small trials
(Smith et al., 2017) Smith and colleagues pooled RCTs comparing exercise where pain was «encouraged or allowed» against pain-free exercise, in adults with musculoskeletal pain lasting more than 3 months. Seven trials met the criteria (385 participants in the pooled short-term analysis, mean age 47; the text calls all programmes home-based, though Table 2 shows a sports-centre arm and regular physiotherapist contact in most trials). Every trial enrolled a localized pain condition (shoulder, Achilles, plantar heel, low back). Pain outcomes, negative SMD favouring painful exercise:
| Horizon | Analysis | n (trials) | SMD (95% CI) | GRADE |
|---|---|---|---|---|
| Short (<=3 mo) | primary | 385 (6) | -0.28 (-0.49 to -0.08) | low |
| Short | sensitivity | 215 (3) | -0.27 (-0.54 to -0.05) | moderate |
| Medium (3-12 mo) | primary | 173 (3) | -0.59 (-1.03 to -0.15) | low |
| Medium | sensitivity | 40 (1) | -0.32 (-0.95 to 0.31) | low |
| Long (>=12 mo) | primary | 345 (5) | 0.01 (-0.39 to 0.41) | very low |
| Long | sensitivity | 215 (3) | 0.13 (-0.14 to 0.40) | moderate |
Source: Table 3. Function and disability were not pooled; narratively, every arm improved, «with no clear superiority» in five of six trials (Rathleff’s plantar-fasciitis trial favoured the loaded arm at 3 months only). «Clinically significant improvements in patient-reported outcome measures were reported across all interventions and control exercises, and all time points», so pain-free programmes also improved. In one shoulder trial (Holmgren) fewer patients in the loaded arm chose surgery «(63%)» control vs «(24%; p<0.0001)», but its comparator was non-specific exercise, so specificity, load and pain allowance are confounded. (Smith et al., 2017)
Reading the table by horizon (inferred from Smith et al., 2017):
- Short term: a small edge for the painful arm. Both analyses agree on about -0.27 to -0.28. The paper gives no minimal clinically important difference; it grades effects by Cohen’s 0.2/0.5/0.8. On a 0-100 mm VAS, using the 21-29 mm SDs the low-back trial reports, -0.27 is roughly 6-8 mm, a small difference in an unblinded trial with a self-reported outcome.
- Medium term: direction favours the painful arm, certainty low. The 3-trial pool is -0.59 and excludes zero; the paper’s «no difference at medium-term» appears to rest on a single-trial sensitivity analysis («The one remaining trial showed no significant difference in the medium term.51»). The source is also internally unclear about which trial that is (it cites ref 51, Rathleff, as both excluded and remaining; probably Rathleff, who randomized 48 — Table 3’s n=40 equals Silbernagel’s randomized n, but Silbernagel is explicitly excluded). This is insufficient evidence either way, not a demonstrated null.
- Long term: no significant difference. Both analyses straddle zero; the moderate-GRADE sensitivity interval (-0.14 to 0.40) rules out a large long-run advantage for either approach.
- No analysis significantly favours pain-free exercise. Every short- and medium-term point estimate sits on the painful side; the two long-term estimates sit just on the pain-free side (0.01, 0.13) with intervals spanning zero.
What “painful” meant — a convention, not a tested dose
(Smith et al., 2017) The trials differed on whether pain was merely allowed or recommended, whether a ceiling was set, and how fast pain had to settle: «immedi- ately», «by the next session», or «by the next day». Four of the seven named a level at about half the scale (up to 50 mm on a 100 mm VAS, or 5/10; a cap in some, a recommended target in Holmgren; Maenhout supplied no data to the pooled analyses), with pain to settle after each set, by the next session, or by the next morning (one of the four adding no morning stiffness) (Table 2). Across all seven, pain had to subside immediately in three (Aasa, Rathleff, Littlewood), by the next session in one (Holmgren) and by the next day in three. «Unfortunately none of the trials included in this review recorded the level of pain patients actually experienced during their exercise programme».
- Studied range, not a located threshold (inferred from Smith et al., 2017): the ~5/10 level is the only numeric convention the trials used. No trial compared it with a higher or lower ceiling, and none measured the pain actually felt, so 5/10 is the edge of the studied range, not a knee in a dose-response curve -> The Underivable Optimum.
What the contrast cannot separate — pain or load
«In addition to the aspect of pain, an important difference between the intervention arm and the control arm is the higher loads, or levels of resistance, employed with the exercises into pain, and it is unknown if the difference in responses can be attributable to these two elements of the different exercise programmes.» (Smith et al., 2017) The comparators were also mixed: low-load motor control, stretching, pain-free cuff work, and in one trial usual physiotherapy including injections (Table 2).
- So the review answers does a programme that allows pain do worse? (not shown to), not does pain itself help? The painful arms were mostly heavy or eccentric loading; the short-term edge may come from the load, the dose, or the instruction. (inferred from Smith et al., 2017)
- The same pattern — an apparent edge between exercise options that may reflect more training stimulus rather than the labelled difference — recurs in Is the Exercise Modality Doing Any Work. (inferred from Smith et al., 2017)
Who it transports to
«One trial included low back pain,43 72 three included shoulder pain,47 52 70 71 two included Achilles pain73 74 and one included plantar heel pain.51» Widespread pain was excluded: «This review excluded trials where participants had a diagnosis of more widespread pain disorders like fibromyalgia.» (Smith et al., 2017)
- Six of seven trials are localized tendon or fascia conditions (shoulder, Achilles, plantar fascia); one is low back pain. No osteoarthritis, neck, or knee trial is included. The result transports to localized, load-sensitive tendon-type pain (the tissue grouping is the wiki’s; the review declined to subgroup by tissue); for fibromyalgia or widespread pain it says nothing. The paper also cites the counter-recommendation, made for chronic musculoskeletal pain generally (so possibly applying within the included population too), that «exercising non-painful body parts with patients with chronic musculoskeletal pain has been recom- mended» (ref 87, after a study showing dysfunctional endogenous analgesia), and counters it by noting that «empirical data within this field are greatly lacking» and that «this systematic review shows that painful exercises may even improve the clinical outcomes». (inferred from Smith et al., 2017)
- Safety was not measured as an outcome. The review reports pain and function/disability only; no
adverse-event, flare or dropout-by-arm analysis [searched:
adverse/flare/side effect= 0 across chunk 01]. The defensible claim is that painful protocols were not shown to produce worse pain or (narratively) function outcomes, not that they are proven harmless. - Allegiance. Two review authors led or co-authored included trials (Rathleff, ref 51; Bateman, ref 52), and review authors wrote two of the papers the review itself cites as proponents of exercising into pain (refs 37, 38); competing interests are declared none. (Smith et al., 2017) This does not change the numbers, but nothing in the corpus independently replicates them.
Decision relevance
(inferred from Smith et al., 2017)
- Do not stop a rehab exercise just because it hurts somewhat. For localized chronic tendon, fascia or
low-back pain, programmes that allowed pain were not shown to do worse than pain-free ones on pain or
(narratively) function at any horizon, and did slightly better on pain in the short term; the long-term
intervals cannot exclude a small (up to ~0.4 SMD) disadvantage. A strict pain-free rule has no outcome support in the
held evidence; its likely cost is a lower load, though the review did not report adherence by arm [searched:
adherence= 0;complianceonce, in the blinding discussion; chunk 01]. - Labelled practice default (a convention from the trials, not a tested threshold), within a prescribed rehab programme for a localized pain condition: pain during the exercise up to about 5/10 that settles soon after the session, with no rise in daily pain or morning stiffness. The trials behind the moderate-certainty short-term result (Aasa, Holmgren, Rathleff) required pain to subside after each set or by the next session; next-morning settling is the most permissive rule any trial used, and those trials are absent from both moderate-certainty analyses. Pain above ~5/10 exceeds the four protocols that named a level (three set none, and no trial recorded pain felt); pain lasting past the next day lies outside every studied protocol (untested, not shown safe or harmful) — reduce the load.
- Pain is allowed, not required. The evidence does not show that seeking pain adds benefit. The painful arms also loaded more (often progressively, e.g. Rathleff from 12RM over 3 months), so what the trials tested is a bundle — heavier loading with some pain tolerated — not pain itself.
- Pain-free programmes also improved. A strict pain-free rule has no support for being better, but someone who will not tolerate pain still has an option whose arms also improved substantially (slightly less on short-term pain, possibly more on medium-term pain at low certainty; comparably on function, narratively), so the stakes of this choice are low.
- Firmness: a cheap, reversible adjustment (reduce load if pain overshoots); no analysis favoured pain-free exercise, so it warrants a plain default despite low-to-moderate certainty -> Acting on a Direction Without a Magnitude. The long-term picture is the honest ceiling: over a year no difference was detected; the moderate-certainty interval (-0.14 to 0.40) excludes a large effect either way but not a small one favouring pain-free exercise.
Limits and gaps
- Small and unblinded: seven trials (six pooled, 385 people); «no trial blinded the participants»; self-reported pain. (Smith et al., 2017)
- Pain ceiling untested — no trial randomized different pain allowances at equal load (
type-G). - Pain actually felt was never recorded, so dose cannot be estimated (
type-G). - Osteoarthritis, neck and widespread pain are absent from the evidence (
type-G). - Adherence by arm was not analysed. The review analysed pain (pooled) and function (narratively)
only and reports no adherence by arm; whether a pain allowance keeps more people exercising (or drives
more to quit) is unknown from the held evidence. The only by-arm datum: in Littlewood, attrition was
«balanced across the intervention and control groups» (weak evidence that pain allowance did not drive
dropout in that one trial). The same in-task-to-
behaviour gap blocks the adherence claim for unpleasant high-intensity exercise
(Affective Response to High-Intensity Interval Exercise) (
type-G). (inferred from Smith et al., 2017) - The loop is open: this grades how faithfully the page represents one review, not whether the default improves anyone’s outcome.