Dose-response effects of transcutaneous electrical nerve stimulation for knee osteoarthritis: A systematic review and meta-analysis

Abstract

Objective

To clarify the analgesic efficacy of transcutaneous electrical nerve stimulation (TENS) in knee osteoarthritis (KOA) and examine whether stimulation parameters moderate treatment effects.

Methods

We meta-analysed randomized controlled trials evaluating TENS for pain in KOA. Thirty-six trials (n = 2518) were included. Pain outcomes were pooled overall and stratified by comparator (control/placebo vs other active treatments). Mixed-effects models examined moderation by frequency, intensity, session duration, and number of sessions.

Results

Across all trials, TENS produced a small but statistically significant reduction in pain. Effects were larger versus control/placebo, suggesting a specific benefit relative to minimal or no intervention. Effects were smaller versus other active treatments, likely reflecting effective co-interventions. Session duration moderated analgesia, with greater pain relief when stimulation lasted ≥40 min. No significant moderation was observed for frequency, intensity, or number of sessions.

Conclusion

TENS may provide clinically relevant pain relief in KOA when delivered with treatment fidelity to pre-specified, literature-informed dosing criteria. However, higher-quality sham-controlled trials with complete parameter reporting and pain assessed during or immediately after stimulation are needed to better quantify the specific analgesic effect beyond contextual influences.

1. Introduction

Osteoarthritis (OA) is a major contributor to disability and healthcare burden [1,2]. OA is incurable, and disease-modifying drugs are not available [3]. In KOA, pain and functional limitations reduce mobility and quality of life [4,5]. Analgesics are commonly used but carry gastrointestinal and cardiovascular risks, particularly in older adults with comorbidities [6,7]. Non-pharmacological management is central, with education, weight loss, and structured exercise recommended as first-line care [8,9] Because pain can limit exercise participation, safe analgesic adjuncts may facilitate adherence [8,10].

Transcutaneous electrical nerve stimulation (TENS) is a non-pharmacological intervention for chronic pain [11,12]. Delivered via surface electrodes, it activates peripheral nerves to modulate nociceptive processing through peripheral and central mechanisms [12,13]. TENS is safe, non-invasive, and can be self-administered after instruction, supporting long-term use in knee osteoarthritis (KOA) [14,15]. Evidence suggests that clinical effects depend on stimulation parameters (e.g., frequency, intensity, session duration) [[16], [17], [18]]. Inadequate control or reporting of key stimulation parameters has been associated with reduced or absent effects in clinical trials [[19], [20], [21], [22], [23]], potentially underestimating true efficacy.

Evidence regarding the efficacy of TENS specifically for KOA remains inconclusive, and major guidelines have recommended against or not included TENS for KOA management [3,8,24]. These positions reflect the inconsistent findings reported in previous systematic reviews and meta-analyses. Prior meta-analyses report mixed findings [[25], [26], [27], [28]], but none accounted for variation in stimulation parameters or comparator type, limiting interpretability. Collectively, these recommendations and mixed findings underscore the need for a synthesis that accounts for stimulation parameters and comparator type when estimating TENS efficacy in KOA.

To address these gaps, we conducted a systematic review and random-effects meta-analysis to estimate the effect of TENS on post-treatment KOA pain, examine comparator type (control/placebo vs active treatments) as a moderator, and explore dose–response patterns using clinically reasoned criteria for frequency, intensity, session duration, and number of sessions. Findings were interpreted in light of risk of bias, heterogeneity and potential small-study effects. Consistent with current guidance, TENS should be appraised as a potential adjunct to core active care (education and exercise), particularly where short-term analgesia could facilitate participation.

2. Methods

This systematic review was designed and reported in accordance with the PRISMA guidelines. The study protocol was prospectively registered in the PROSPERO database on November 1, 2024 (CRD42024589441). After protocol registration, we focused the quantitative synthesis exclusively on pain intensity, the primary therapeutic target of TENS and the most consistently reported outcome across trials. Accordingly, other outcomes (e.g., function) were not synthesized. When multiple time-points were available, we prioritized the assessment closest to end-of-treatment (or during/immediately after stimulation when available) to approximate peak analgesia. Comparator-stratified and dosing analyses were planned as secondary/exploratory.

2.1. Search strategy

A comprehensive search was conducted in MEDLINE (via PubMed), the Cochrane Central Register of Controlled Trials, and Scopus from inception to November 17, 2024. Reference lists were screened, and searches combined controlled vocabulary and keywords for KOA and TENS. Only studies involving human participants and published in English, Spanish, Portuguese, or French were considered. Detailed search strategies are provided in Supplementary File S1.

2.2. Eligibility criteria and selection

Studies were included if they were RCTs in adults (≥18 years) with clinically or imaging-diagnosed KOA; evaluated TENS (alone or combined with other treatments); reported pain intensity from baseline to follow-up using a validated quantitative scale; and compared TENS with no intervention, placebo, or active treatments. Co-interventions were permitted if applied equally across groups. Studies using invasive electrical techniques (e.g., electroacupuncture, percutaneous electrical nerve stimulation) or non-sensory modalities (e.g., microcurrent stimulation) were excluded. Duplicate records were removed, and two reviewers (SLB and JJL) independently screened titles/abstracts and assessed full texts when needed; disagreements were resolved by discussion or a third reviewer (JJAC).

2.3. Data extraction

A standardized extraction form was used to collect data on study characteristics, including trial design, participants, comparators, TENS parameters, adverse events, and pre/post-intervention pain outcomes. If both resting and movement-related pain data were available, resting pain scores were extracted, as these were the most consistently reported across studies. For trials reporting multiple time-points, we prioritized the assessment closest to end-of-treatment, or during/immediately after stimulation when available, to approximate peak TENS analgesia.

To evaluate dose-related effects, four key TENS parameters were extracted from each study and categorized as “meets pre-specified criteria,” or “does not meet pre-specified criteria” based on pre-specified, literature-informed treatment-fidelity criteria.

• 1.Frequency >10 Hz (up to 200 Hz) was classified as meets pre-specified criteria, whereas ≤10 Hz was classified as does not meet pre-specified criteria [[29], [30], [31], [32], [33], [34]]. Mixed-frequency protocols alternating between low and high frequencies were coded as mixed/alternating and classified as meeting the pre-specified frequency criterion, consistent with prior proposals that varying stimulation parameters may mitigate analgesic tolerance [29,32]; accordingly, they were not forced into a single low- or high-frequency category.
• 2.Intensity sufficient to produce a sustained sensory perception, ranging from a distinct but comfortable sensation to the maximum tolerable level [35,36]. Intensity was classified as does not meet pre-specified criteria if it: (1) triggered motor activation which was treated as a distinct delivery feature (not evidence of harm) and coded as does not meet pre-specified criteria for sensory-level intensity; (2) was fixed without individual calibration; or (3) restricted participant-driven adjustment during sessions, which is necessary to counteract habituation.
• 3.Session duration of at least 40 min, based on evidence suggesting this threshold yields optimal analgesic outcomes in KOA. Cheing et al. (2003) demonstrated that a 40-min TENS session yielded greater and longer-lasting analgesia than shorter sessions [37].
• 4.A minimum of 10 treatment sessions was treated as an operational, pre-specified exposure threshold, based on evidence supporting cumulative effects in chronic pain conditions [16,38,39].

When data were unclear or missing, corresponding authors were contacted for clarification. Two independent reviewers (SLB and JJL) extracted the data, resolving disagreements by discussion or consultation with a third reviewer (JJAC) when necessary.

2.4. Assessment of methodological quality and bias risk

Two reviewers (MDA and FJFS) independently assessed study quality using the PEDro scale [40] and version 2 of the Cochrane risk-of-bias tool for randomized trials (RoB 2) [41]. The PEDro scale comprises 11 items evaluating key trial-quality domains [40]. Only items 2–11 contribute to the internal validity score (0–10 points), while item 1 pertains to external validity and is not scored. Studies scoring ≥6 were categorized as moderate to high quality. Discrepancies were resolved through discussion.

Risk of bias was assessed using the RoB 2 tool [41]. Five domains were assessed (D1–D5: randomization, deviations from intended interventions, missing outcome data, outcome measurement, and selection of the reported result) [41]. For each domain, a judgment was assigned as “Low risk” to indicate minimal bias, “High risk” for substantial potential bias, or “Some concerns” when insufficient detail was available to determine the level of risk. Risk-of-bias judgements were summarised at trial level with attention to pain outcomes; domain-level summaries are shown in the main figures.

2.5. Data synthesis

Effect sizes (d) were computed as standardized mean differences in change scores between TENS and control groups [42]. For each study and outcome (k), d was calculated by subtracting the control group's mean change score from that of the TENS group, divided by the pooled standard deviation of the pre-intervention scores. Small-sample bias was corrected using a standard adjustment. Variance of effect sizes was estimated using the method described by Morris (2008) [42]. In the absence of reported Pearson correlation coefficient between pretest and posttest/follow-up scores, a value of r = 0.7 was imputed following Rosenthal (1991) [43]. Positive d values indicate greater improvement in the TENS group; thresholds were 0.20/0.50/0.80 (small/moderate/large) [44].

When multiple assessment methods were used for the same outcome, individual effect sizes were averaged in several trials [45,46]; for studies with two intervention groups and one control, the control sample size was split to avoid statistical dependence [37,[47], [48], [49]]; for trials with one intervention group and two control groups, the intervention group sample size was halved [10,[50], [51], [52], [53], [54], [55], [56], [57], [58], [59]]; and in multi-arm trials, sample sizes were divided by the number of relevant groups [45].

2.6. Statistical analysis

A random-effects model was used. The primary analysis pooled all eligible trials across comparators (random-effects). Comparator-stratified results (control/placebo vs active treatments) are reported as a secondary analysis, and dosing analyses were exploratory. Standardized mean change differences were weighted by inverse variance, and between-study variance (τ²) was estimated using restricted maximum likelihood [60]. Pooled effect sizes (d+) and 95 % confidence intervals were calculated using the Hartung-Knapp adjustment [61]. Statistical significance was determined using the t-distribution. Cochran's Q-statistic and the I² index were calculated to assess variability among effect sizes, with Q-statistic p ≤ 00.05 indicating variability and I² index thresholds for interpreting heterogeneity [62,63] as follow: <25 % (none), 25–49.9 % (low), 50–74.9 % (moderate), and >75 % (high). Forest plots were used for visual inspection.

Outlier and influence diagnostics were used to assess robustness [64,65]; potential outliers were flagged using studentized deleted residuals (|rstudent| > 1.96), and influential cases were identified using case-deletion diagnostics (details in Supplementary File S3; Supplementary Fig. S1–S2).

Moderator analyses were conducted to examine whether stimulation frequency, intensity, session duration, number of sessions, and comparator type moderated effect sizes. Categorical moderators were tested using mixed-effects, inverse-variance–weighted ANOVAs with the Hartung–Knapp approach [66,67]. Variance explained by moderators was estimated [68], and model misspecification was assessed using Q_E. Studies coded as not classifiable (NR/insufficient reporting) for a given parameter were excluded from the corresponding parameter-specific moderator analyses, while remaining eligible for the overall meta-analysis.

Finally, publication bias was evaluated via funnel plots, Egger's test [69], and the trim-and-fill method [70]. All analyses were conducted in R using the metafor package, with a two-tailed significance level set at p ≤ 00.050.

3. Results

The search identified 1897 records (1893 from databases and four from reference lists). After de-duplication, 1105 records were screened, 140 full texts were assessed, and 36 studies were included (Fig. 1).

Fig. 1.

PRISMA (2020) flow diagram. PRISMA, preferred reporting items for systematic reviews and meta-analyses.

Across the included studies, 2518 participants were analysed (1116 received TENS), with a mean age of 62.5 years and 77.9 % women (Table 1), consistent with KOA epidemiology [25,71].

Table 1.

Characteristics of the included studies.

Author, Year	Design	Interventions Groups	Sample Size per Group	Female (% sample)	Age ± SD	Outcome (Ranges)	Pain score during or immediately after TENS	Baseline Pain Intensity ± SD
Alcidi et al., 2007 [72]	RCT	TENS/	20/	80/	69.4 ± 9.2/	Visual analogue scale (0–100)	Yes	53.85 ± 25.70/
		Radiofrequency radiation	20	90	62 ± 10.9			63.85 ± 20.82
Altaş & Demirdal, 2020 [73]	RCT	TENS + Exercise/	20/	70/	57.6 ± 8.7/	Visual analogue scale (0–10)	No	7.3 ± 1.0/
		Exercise	20	85	55.5 ± 9.6			7.3 ± 1.1
Artuç et al., 2023 [50]	RCT	Hot Pack + TENS/	20/	85/	58.50 ± 9.80/	Visual analogue scale (0–10)	No	7.80 ± 1.39/
		Hot Pack + IF/	20/	80/	61.95 ± 11.78/			7.65 ± 1.22/
		Hot Pack + Placebo TENS/	20/	85/	56.05 ± 9.45/			7.90 ± 1.71/
		Hot Pack + Placebo IF	20	85	54.00 ± 8.79			7.60 ± 1.60
Atamaz et al., 2012 [51]	RCT	TENS/	37/	83.8/	61.9 ± 6.9/	Visual analogue scale (0–100)	No	76.1 ± 16.2/
		Placebo TENS/	37/	73.0/	60.7 ± 6.5/			74.4 ± 16.4/
		IF/	31/	87.1/	62.0 ± 7.9/			74.4 ± 16.7/
		SWD	31	87.1	61.6 ± 7.4			78.5 ± 15.6
Burch et al., 2007 [74]	RCT	TENS/	55/	78.2/	60.8 ± 11.4/	Visual analogue scale (0–100)	No	9.6 ± 3.5/
		IF + Patterned muscle stimulation	54	66.7	62.6 ± 10.5			9.6 ± 3.3
Cetin et al., 2008 [52]	RCT	TENS + Hot Pack + Exercise/	20/	100/	61.85 ± 8.64/	Visual analogue scale (0–10)	No	5.85 ± 1.34/
		SWD + Hot Pack + Exercise/	20/	100/	59.75 ± 11.63/			5.69 ± 1.55/
		Ultrasound + Hot Pack + Exercise/	20/	100/	57.60 ± 7.33/			5.90 ± 1.45/
		Hot Pack + Exercise	20	100	61.05 ± 8.26			5.76 ± 1.48
Chaturvedi et al., 2021 [53]	RCT	TENS + Placebo tDCS/	18/	66.66/	54.78 ± 5.91/	Visual analogue scale (0–10)	No	6.01 ± 1.39/
		Placebo TENS + Placebo tDCS/	18/	77.77/	51.75 ± 5.79/			5.55 ± 1.12/
		Placebo TENS + tDCS	18	61.11	52.78 ± 5.37			5.51 ± 1.13
Cheing et al., 2003 [37]	RCT	TENS 20 min/	10/	90/	69.2 ± 5.7/	Visual analogue scale (0–10)	No	4.4 ± 1.2/
		TENS 40 min/	10/	90/	63.2 ± 8.4/			5.2 ± 1.1/
		TENS 60 min/	10/	90/	63.5 ± 5.6/			5.5 ± 1.3/
		Placebo TENS	8	87.5	66.1 ± 9.3			4.9 ± 1.0
Cherian et al., 2015 [75]	RCT	TENS/	13/	84.6/	54 ± 9/	Visual analogue scale (0–10)	No	4.95 ± 2.25/
		Exercise and/or corticosteroid injections	10	70	55 ± 12			4.50 ± 2.50
Coelho de amorim et al., 2014 [76]	RCT	TENS/	12/	75/	66.5 ± 11.35/	Visual analogue scale (0–10)	Yes	5.4 ± 2.1/
		Manual therapy	12	91.6	67.9 ± 15.2			5 ± 2.2
Elbadawy, 2016 [77]	RCT	TENS/	30/	66.6/	59.93 ± 4.35/	Visual analogue scale (0–10)	No	7.49 ± 0.79/
		Periosteal stimulation therapy	30	66.6	59.43 ± 4.17			7.71 ± 0.76
Iijima et al., 2020 [78]	RCT	TENS/	30/	76.7/	59.9 ± 6.41/	Visual analogue scale (0–100)	Yes	16.4 ± 18.9/
		Placebo TENS	29	69	58.2 ± 5.63			15.1 ± 15.8
İnal et al., 2016 [47]	RCT	HF TENS/	30/	100/	64.1 ± 0.99/	Visual analogue scale (0–10)	No	8.97 ± 1.92/
		LF TENS/	30/	100/	64.4 ± 1.70/			8.8 ± 2.03/
		Placebo TENS	30	100	64.6 ± 1.88			8.7 ± 1.35
Isik et al., 2016 [79]	RCT	TENS/	44/	81.8/	53.8 ± 12.6/	Visual analogue scale (0–100)	Yes	59.0 ± 19.0/
		Medicinal leech	46	95.6	59.6 ± 8.8			66.0 ± 16.5
Kędzierski et al., 2012 [80]	RCT	TENS/	25/	NR/	70.4 ± 11.6/	Visual analogue scale (0–10)	No	6.3 ± 2/
		Laser therapy	25	NR	70.2 ± 11.1			6 ± 1.8
Law & Cheing, 2004 [48]	RCT	LF TENS/	13/	100/	82.7 ± 6.1/	Visual analogue scale (0–10)	Yes	6.6 ± 2.0/
		HF TENS/	12/	100/	84.3 ± 6.9/			5.2 ± 1.8/
		LF-HF TENS/	13/	92.3/	80.00 ± 5.8/			5.4 ± 2.2/
		Placebo TENS	10	100	83.2 ± 5.4			5.8 ± 3.0
Maheu et al., 2022 [81]	RCT	TENS/	55/	67.3/	66.9 ± 8.1/	Numerical rating scale (0−10)	No	5.9 ± 1.3/
		Weak opioids	55	61.8	66.0 ± 7.8			5.8 ± 1.3
Mahmood et al., 2017 [82]	RCT	TENS/	27/	NR/	NR/	Visual analogue scale (0–10)	No	7.41 ± 2.26/
		Control	28	NR	NR			8.21 ± 2.32
Mascarin et al., 2012 [45]	RCT	TENS/	12/	100/	64.8 ± 7.0/	Visual analogue scale (0–10)	No	5.6 ± 2.7/
		Kinesiotherapy/	16/	100/	59.6 ± 7.2/			7.0 ± 2.1/
		Ultrasound	12	100	62.8 ± 7.6			7.3 ± 2.3
Mukharjee & Prabhakar, 2020 [83]	RCT	TENS + Exercise/	10/	60/	NR/	Visual analogue scale (0–10)	No	5.90 ± 1.20/
		IF + Exercise	10	70	NR			6.50 ± 0.85
Mutlu et al., 2018 [54]	RCT	TENS/	22/	86.4/	57.77 ± 6.24/	Visual analogue scale (0–10)	No	5.77 ± 3.66/
		Mobilization with movements/	21/	100/	54.19 ± 7.34/			6.90 ± 3.04/
		Passive joint mobilization	21	76.2	57.77 ± 6.24			3.95 ± 3.91
Ng et al., 2003 [84]	RCT	TENS/	8/	NR/	85.88 ± 5.96/	Visual analogue scale (0–10)	Yes	4.19 ± 1.49/
		Electroacupuncture	8	NR	84.38 ± 6.48			4.69 ± 1.71
Paker et al., 2006 [90]	RCT	TENS/	30/	NR/	54.18 ± 8.19/	WOMAC pain (0–20)	Yes	9.55 ± 3.87/
		Hyaluronic acid injection	30	NR	64.04 ± 8.62			11.20 ± 3.59
Palmer et al., 2014 [55]	RCT	TENS + Exercise/	73/	64.4/	61.2 ± 11.4/	WOMAC pain (0–20)	No	9.0 ± 6.0/
		Placebo TENS + Exercise/	74/	66.2/	60.9 ± 10.8/			9.0 ± 5.0/
		Exercise	77	58.4	62.0 ± 9.4			8.0 ± 5.8
Pietrosimone et al., 2009 [56]	RCT	TENS/	10/	40/	56 ± 10.1/	Visual analogue scale (0–100)	Yes	17.05 ± 15.69/
		Focal joint cooling/	11/	45.5/	58 ± 8.4/			19.23 ± 15.56/
		Control	12	58.3	54 ± 9.9			19.83 ± 15.59
Pietrosimone et al., 2010 [57]	RCT	TENS/	12/	50/	60.3 ± 1.9/	Visual analogue scale (0–10)	No	8.2 ± 12.9/
		Placebo TENS/	11/	72.7/	58.7 ± 12.2/			10.1 ± 6.7/
		Control	12	58.3	58.3 ± 11.8			5.3 ± 7.0
Pietrosimone et al., 2011 [58]	RCT	TENS + Exercise/	12/	50/	NR/	WOMAC pain (5−25)	No	14.7 (11.4–18.1)a/
		Placebo TENS + Exercise/	12/	66.6/	NR/			15.0 (11.7–18.6)a/
		Exercise	12	58.3	NR			14.3 (11.0–17.0)a
Pietrosimone et al., 2020 [10]	RCT	TENS + Exercise/	32/	56.3/	60.8 ± 7.3/	WOMAC pain (5–100)	No	43.8 ± 20.0/
		Placebo TENS + Exercise/	29/	65.5/	62.5 ± 7.7/			42.7 ± 21.6/
		Exercise	29	48.3	63 ± 7.4			40.2 ± 15.4
Po-En Chiu et al., 2022 [85]	RCT	TENS/	16/15	62.5/	62.81 ± 5.72/	Visual analogue scale (0–10)	Yes	5.81 ± 0.91/
		Dry needling		73.3	65.73 ± 6.79			5.80 ± 1.42
Reichenbach et al., 2021 [86]	RCT	TENS/	108/	48/	65/	Visual analogue scale (0–10)	No	4.1 ± 1.7/
		Placebo TENS	112	54	66			4.1 ± 2.1
Sajadi et al., 2020 [87]	RCT	TENS/	20/	95/	65.85 ± 5.81/	Visual analogue scale (0–100)	No	71.00 ± 15.18/
		tDCS	20	75	59.30 ± 6.13			68.50 ± 19.54
Sangtong et al., 2019 [88]	RCT	TENS + Ultrasound/	74/	93.2/	63.4 ± 7.6/	Numerical rating scale (0−10)	Yes	3.9 ± 2.0/
		Ultrasound	74	89.2	62.5 ± 8.0			3.9 ± 2.1
Shimoura et al., 2019 [46]	RCT	TENS/	25/	76.0/	59.1 ± 6.13/	Visual analogue scale (0–10)	Yes	1.57 ± 1.67/
		Placebo TENS	25	64.0	57.9 ± 5.07			1.89 ± 2.08
Vance et al., 2012 [49]	RCT	HF TENS/	25/	56/	57 ± 11.8/	Visual analogue scale (0–100)	No	17.7 ± 5.0/
		LF TENS/	25/	64/	55 ± 14.4/			29.5 ± 5.3/
		Placebo TENS	25	64	57 ± 10.9			21.2 ± 5.1
Wen-Ling Chen et al., 2013 [89]	RCT	TENS/	23/	89.9/	66.52 ± 7.20/	Visual analogue scale (0–10)	No	6.11 ± 1.37/
		Hyaluronic acid injection	27	85.2	67.96 ± 9.94			6.46 ± 1.82
Yurtkuran & Kocagil, 1999 [59]	RCT	TENS/	25/	92/	NR/	Present pain intensity, Likert scale (0–5)	No	1.2 ± 1.1/
		Ice massage/	25/	100/	NR/			0.7 ± 0.6/
		Electropuncture/	25/	84/	NR/			1.36 ± 0.5/
		Placebo TENS	25	88	NR			0.8 ± 0.5

RCT: randomized controlled trial; IF: interferential current; SWD: shortwave diathermy; tDCS: transcranial direct current stimulation; PT: physiotherapy; NR: not reported; HF: high frequency; LF: low frequency; WOMAC: Western Ontario and McMaster Universities Osteoarthritis Index.

^aMinimum and maximum values within the data range.

Pain was assessed using VAS/NRS in 31/36 trials [37,[45], [46], [47], [48], [49], [50], [51], [52], [53], [54],56,57,59,[72], [73], [74], [75], [76], [77], [78], [79], [80], [81], [82], [83], [84], [85], [86], [87], [88], [89]]; WOMAC pain in four trials [10,55,58,90]; and a Likert Present Pain Intensity scale in one trial [59].

Across trials, baseline pain intensity averaged 5.52/10 (range 0.96–9.0), consistent with moderate pain. Although most trials enrolled participants with moderate baseline pain, six trials reported mean baseline pain <3/10 (or <3/10-equivalent after rescaling when other scales were used) [46,49,56,59,74,78], potentially limiting room for improvement (floor effects). Because only six trials met this threshold, we did not perform a sensitivity analysis based on baseline pain severity. No included trial reported statistically significant between-group differences in baseline pain intensity, indicating comparable pain levels across arms before treatment. (Table 1).

Details of TENS delivery and parameter-based classification are summarised in Table 2 and Supplementary File S2. Briefly, most trials used stimulation >10 Hz and sensory-level intensity, while only 12/36 delivered ≥40 min per session and 25/36 delivered ≥10 sessions. No serious adverse events were reported among trials providing adverse-event data.

Table 2.

The TENS parameters applied in each included trial.

Author, Year	Frequency	Intensity described	Session duration	Number of sessions	Main findings in terms of pain
Alcidi et al., 2007 [72]	50 Hz ✓	Well-tolerated and stable sensation of tingling. ✓	20 min ✗	5 sessions over 1 week. ✗	TENS group showed no significant reduction (p > 0.05), while the radiofrequency radiation group showed significant improvement (p < 0.001).
Altaş & Demirdal, 2020 [73]	0–100 Hz ✓	Strong but tolerable. ✓	20 min ✗	10 sessions over 2 weeks. ✓	TENS group showed a significant reduction vs. control (p < 0.05).
Artuç et al., 2023 [50]	80 Hz ✓	Tingling sensation. ✓	20 min ✗	10 sessions over 3 weeks. ✓	TENS produced a significant reduction (p < 0.001), with no significant differences between groups (p > 0.05).
Atamaz et al., 2012 [51]	80 Hz ✓	Set between 10 and 30 mA ✗	20 min ✗	15 sessions over 4 weeks. ✓	TENS group showed a significant reduction (p < 0.05), with no significant differences between groups (p > 0.05).
Burch et al., 2007 [74]	0.2 Hz ✗	Perceived, without muscular contraction. ✓	35 min ✗	56 sessions over 8 weeks. ✓	TENS group showed no significant reduction (p > 0.05), while the IF + patterned muscle stimulation group showed significant between-group improvement (p = 0.004).
Cetin et al., 2008 [52]	60–100 Hz ✓	Comfortable, without muscular contraction. ✓	20 min ✗	24 sessions over 8 weeks. ✓	TENS group showed a significant reduction (p < 0.0001), with no significant differences between groups (p > 0.05).
Chaturvedi et al., 2021 [53]	100 Hz ✓	NR	20 min ✗	20 sessions over 6 weeks. ✓	TENS group showed a significant reduction (p < 0.015) compared to control.
Cheing et al., 2003 [37]	100 Hz ✓	Strong but comfortable tingling paresthesia. ✓	40 min ✓	10 sessions over 2 weeks. ✓	Longer stimulation duration, such as 40 or 60 min of TENS produced a more long-lasting analgesic effect than 20 min of TENS (p < 0.05).
Cherian et al., 2015 [75]	NR	NR	Used all day, except bathing. ✓	Daily use for 12 weeks. ✓	TENS group showed a significant reduction (p = 0.0027) compared to control.
Coelho de amorim et al., 2014 [76]	80 Hz ✓	Perceived as comfortable. ✓	20 min ✗	12 sessions over 4 weeks. ✓	TENS produced a significant reduction (p < 0.05), with no significant differences with manual therapy (p > 0.05).
Elbadawy, 2016 [77]	100 Hz ✓	Strong but comfortable paresthesia. ✓	30 min ✗	10 sessions over 10 weeks. ✓	TENS group showed a significant reduction (p = 0.000), while the Periosteal stimulation therapy group showed significant between-group improvement (p = 0.000).
Iijima et al., 2020 [78]	1–100 Hz ✓	Strong but comfortable, non-painful tingle. ✓	60 min ✓	Single session. ✗	TENS group showed no significant reduction (p > 0.05); compared with placebo TENS.
İnal et al., 2016 [47]	3–100 Hz ✓ 2–4 Hz ✗	Strong but comfortable and tolerable level. ✓	20 min ✗	10 sessions over 2 weeks. ✓	HF and LF TENS group showed a significant reduction (p < 0.05), with no significant differences between groups (p > 0.05).
Isik et al., 2016 [79]	40–150 Hz ✓	Low intensity, barely perceptible. ✗	20 min ✗	15 sessions over 3 weeks. ✓	TENS produced a significant reduction (p < 0.001), with no significant differences with leech therapy (p = 0.085).
Kędzierski et al., 2012 [80]	10 Hz ✗	NR	30 min ✗	10 sessions over 2 weeks. ✓	TENS produced a significant reduction (p < 0.01), while the Laser group showed significant between-group improvement (p < 0.05).
Law & Cheing, 2004 [48]	100 Hz ✓ 2 Hz ✗ 3–100 Hz ✓	Comfortable level determined by participants. ✓	40 min ✓	10 sessions over 2 weeks. ✓	TENS groups showed significant within-group reductions (p = 00.000); no between-group differences but all active TENS outperformed placebo (p < 0.0125)
Maheu et al., 2022 [81]	2–100 Hz ✓	Tingling + muscular contraction. ✗	20 min ✗	NR	TENS produced a significantly greater reduction than the weak opioid group (p = 0.0124).
Mahmood et al., 2017 [82]	NR	NR	15 min ✗	18 sessions over 6 weeks. ✓	TENS group showed a significant reduction (p = 0.002), with no significant differences between groups (p > 0.05).
Mascarin et al., 2012 [45]	100 Hz ✓	Stimulation at sensory threshold. ✗	20 min ✗	24 sessions over 12 weeks. ✓	TENS group showed a significant reduction (p = 0.001), with no significant differences between groups (p > 0.05).
Mukharjee & Prabhakar, 2020 [83]	40–150 Hz ✓	Visible phasic muscle contraction. ✗	10–20 min ✗	10 sessions over 10 days. ✓	TENS group showed no significant reduction (p = 0.06), while the interferential therapy group showed significant improvement (p < 0.001).
Mutlu et al., 2018 [54]	110 Hz ✓	NR	20 min ✗	10 sessions over NR days. ✓	TENS group showed smaller reduction vs manual therapy groups (p = 0.007–0.02); within-group data NR.
Ng et al., 2003 [84]	2 Hz ✗	Visible phasic muscle contraction. ✗	20 min ✗	8 sessions over 2 weeks. ✗	TENS group showed a significant reduction (p < 0.001) compared to control.
Paker et al., 2006 [90]	150 Hz ✓	NR	20 min ✗	15 sessions over 3 weeks. ✓	TENS group showed a significant reduction (p < 0.0001); between-group differences NR.
Palmer et al., 2014 [55]	110 Hz ✓	Strong but comfortable tingling sensation. ✓	As much as needed. ✓	Daily use for 6 weeks. ✓	TENS group showed a significant reduction (p < 0.05) but there were no differences between trial arms (p > 0.05).
Pietrosimone et al., 2009 [56]	150 Hz ✓	Highest tolerable sensory stimulation without contraction. ✓	45 min ✓	Single session. ✗	TENS group showed a significant reduction (p < 0.05). But there were no significant changes between groups (p > 0.05).
Pietrosimone et al., 2010 [57]	150 Hz ✓	Strong sensory stimulation. ✓	At least 8 h a day. ✓	12 sessions over 4 weeks. ✓	TENS group showed no significant difference vs control (p = 0.99) or placebo (p = 0.09).
Pietrosimone et al., 2011 [58]	150 Hz ✓	Strong but comfortable sensory stimulation. ✓	At least 8 h a day. ✓	12 sessions over 4 weeks. ✓	TENS group showed a significant reduction (p < 0.001). But there were no significant changes between groups (p = 0.41).
Pietrosimone et al., 2020 [10]	150 Hz ✓	Strong, manageable sensory stimulation. ✓	Used daily during sessions and daily activities. ✓	10 sessions over 4 weeks. ✓	TENS group demonstrated significantly less pain (P = 0.01) compared to control.
Po-En Chiu et al., 2022 [85]	110 Hz ✓	NR	20 min ✗	Single session. ✗	TENS group showed a significant reduction (p < 0.05)
Reichenbach et al., 2021 [86]	NR	Limit of tolerance, allowing motor threshold. ✗	Up to 60 min ✓	9 sessions over 3 weeks. ✗	TENS group showed no significant difference vs placebo group (p = 0.74).
Sajadi et al., 2020 [87]	100 Hz ✓	NR	25 min ✗	6 sessions over 2 weeks. ✗	TENS and tDCS groups showed no significant differences over time (p > 0.05)
Sangtong et al., 2019 [88]	32–50 Hz ✓	Tolerable tingling sensation. ✓	10 min ✗	6 sessions over 2 weeks. ✗	TENS group showed no significant reduction (p = 0.771)
Shimoura et al., 2019 [46]	1–250 Hz ✓	Strong but comfortable, nonpainful tingle. ✓	60 min ✓	Single session. ✗	TENS group showed a significant reduction (p = 0.030)
Vance et al., 2012 [49]	100 Hz ✓ 4 Hz ✗	NR	40–50 min ✓	Single session. ✗	HF-TENS and LF-TENS increased pain threshold (p ≤ 0.05); only HF-TENS differed from placebo (p = 0.026).
Wen-Ling Chen et al., 2013 [89]	3–20 Hz ✓	From stroking sensation to visible contraction. ✗	20 min ✗	12 sessions over 4 weeks. ✓	TENS group showed a significant reduction (p < 0.001)
Yurtkuran & Kocagil, 1999 [59]	4 Hz ✗	Muscle contraction. ✗	20 min ✗	10 sessions over 2 weeks. ✓	TENS group showed a significant reduction (p < 0.001)

✓, Meets pre-specified criteria; ✗, Does not meet pre-specified criteria; NR, not reported. NR/insufficient reporting indicates that the parameter was not reported or was insufficiently described to allow classification.; HF: high frequency; LF: low frequency.

3.1. Methodological quality and risk of bias assessment

3.1.1. The PEDro scale

The methodological quality of the 36 studies was assessed using the PEDro scale (Table 3). The item that presented the lowest quality was the therapists’ blinding, with all but two studies failing to meet this criterion. [49,50]. Twenty-three studies scored 6–10/10 [10,45,47,48,[50], [51], [52], [53], [54], [55], [56], [57], [58],76,77,79,81,[86], [87], [88], [89], [90]], and thirteen scored ≤5/10 [37,46,59,[72], [73], [74], [75],78,80,[82], [83], [84], [85]].

Table 3.

Assessment of the quality of the studies based on the PEDro scale.

Items
	1	2	3	4	5	6	7	8	9	10	11	Total
Alcidi et al., 2007 [72]	1	1	0	1	0	0	0	0	0	1	1	4
Altaş & Demirdal, 2020 [73]	1	1	0	1	0	0	1	0	0	1	1	5
Artuç et al., 2023 [50]	1	1	0	1	0	0	1	1	0	1	1	6
Atamaz et al., 2012 [51]	0	1	1	1	1	1	1	1	1	1	1	10
Burch et al., 2007 [74]	1	1	1	0	0	0	0	1	0	1	1	5
Cetin et al., 2008 [52]	1	1	0	1	0	0	1	1	0	1	1	6
Chaturvedi et al., 2021 [53]	0	1	1	1	0	0	0	1	1	1	1	7
Cheing et al., 2003 [37]	1	1	0	1	0	0	0	1	0	1	1	5
Cherian et al., 2015 [75]	0	1	0	0	0	0	0	1	0	1	1	4
Coelho de amorim et al., 2014 [76]	1	1	0	1	1	0	1	1	0	1	1	7
Elbadawy, 2016 [77]	1	1	1	1	0	0	1	1	0	1	1	7
Iijima et al., 2020 [78]	0	1	0	0	0	0	0	1	0	1	1	4
İnal et al., 2016 [47]	0	1	1	1	1	0	1	0	0	1	1	7
Isik et al., 2016 [79]	1	1	1	1	0	0	1	1	0	1	1	7
Kędzierski et al., 2012 [80]	0	1	0	1	0	0	0	0	0	1	1	4
Law & Cheing, 2004 [48]	0	1	0	1	1	0	1	1	0	1	1	7
Maheu et al., 2022 [81]	1	1	0	1	0	0	1	1	0	1	1	6
Mahmood et al., 2017 [82]	1	1	0	0	0	0	0	1	0	1	1	4
Mascarin et al., 2012 [45]	1	1	0	1	0	0	1	1	0	1	1	6
Mukharjee & Prabhakar, 2020 [83]	1	1	0	1	0	0	0	1	0	1	1	5
Mutlu et al., 2018 [54]	1	1	1	1	0	0	1	1	0	1	1	7
Ng et al., 2003 [84]	1	1	0	1	0	0	1	0	0	0	1	4
Paker et al., 2006 [90]	0	1	1	1	0	0	1	1	0	1	0	6
Palmer et al., 2014 [55]	1	1	1	1	0	0	1	1	1	1	1	8
Pietrosimone et al., 2009 [56]	1	1	1	1	0	0	1	0	0	1	1	6
Pietrosimone et al., 2010 [57]	1	1	0	1	1	0	1	1	0	1	1	7
Pietrosimone et al., 2011 [58]	1	1	1	1	0	0	1	1	0	1	1	7
Pietrosimone et al., 2020 [10]	1	1	1	1	0	0	1	1	1	1	1	8
Po-En Chiu et al., 2022 [85]	1	1	0	1	0	0	1	0	0	1	1	5
Reichenbach et al., 2021 [86]	1	1	1	1	1	0	1	1	1	1	1	9
Sajadi et al., 2020 [87]	1	1	1	1	0	0	1	1	0	1	1	7
Sangtong et al., 2019 [88]	1	1	1	1	0	0	0	1	1	1	1	7
Shimoura et al., 2019 [46]	1	1	0	1	1	0	0	0	0	1	1	5
Vance et al., 2012 [49]	1	1	1	1	1	1	1	0	0	1	1	8
Wen-Ling Chen et al., 2013 [89]	1	1	0	1	0	0	1	1	0	1	1	6
Yurtkuran & Kocagil, 1999 [59]	1	1	0	0	0	0	1	1	0	1	0	4

1: participant choice criteria are specified; 2: random assignment of participants to groups; 3: hidden assignment; 4: groups were similar at baseline; 5: all participants were blinded; 6: all therapists were blinded; 7: all evaluators were blinded; 8: measurement of at least one of the key outcomes was obtained from more than 85 % of baseline participants; 9: intention-to-treat analysis was performed; 10: results from statistical comparisons between groups were reported for at least one key outcome; 11: the study provides point and variability measures for at least one key outcome.

3.1.2. RoB-2 assessment

RoB 2 indicated that overall risk of bias was high in 47 % of trials, driven mainly by Domain 2 (deviations from intended interventions) and, to a lesser extent, Domain 4 (measurement of the outcome). Domain-level judgements are summarised in Fig. 2, Fig. 3 and in Supplementary File S3; interrater reliability was high (k = 0.814).

Fig. 2.

Summary results of the Risk of Bias 2 (RoB 2) tool assessment.

Fig. 3.

Risk of bias graph according to the Risk of Bias 2 (RoB 2) tool assessment.

3.2. Effects of TENS on pain

Fig. 4 presents the forest plot of the meta-analysis for the pooled standardized mean change difference in pain. Pooling all studies, the overall mean effect size was not statistically significant (d+ = 0.169, SE = 0.093, 95 % CI [−0.017, 0.355], p = 0.074, k = 58). Heterogeneity was considerable (Q(57) = 260.0112, p < 0.0001, I² = 83.9 %). Given this between-study heterogeneity, the overall estimate should be interpreted cautiously and alongside the secondary comparator-stratified and exploratory dosing findings reported below.

Fig. 4.

Forest plot of the meta-analysis for assessing the efficacy of TENS to alleviate pain. RE, random-effects.

3.2.1. Analysis of outliers and influence cases

Two effect sizes were flagged as potential outliers (Altaş & Demirdal, 2020 [73], d = 2.145; Elbadawy et al., 2016 [77], d = 1.681) based on studentized deleted residuals (|rstudent| > 1.96). Case-deletion diagnostics indicated that both were influential (Supplementary Fig. S1–S2; Supplementary File S4). We therefore report the all-studies estimate and a pre-specified sensitivity excluding these two trials. After excluding these cases, the pooled effect was small but statistically significant (d+ = 0.161, SE = 0.075, 95%CI [0.011,0.312], p = 00.036, k = 56). Heterogeneity remained substantial (Q(55) = 164.299, p < 0.0001, I² = 71.8 %).

3.2.2. Effects of TENS dosage on pain

Mixed-effects ANOVAs were conducted to examine whether the effectiveness of TENS on pain was moderated by stimulation parameters, frequency, intensity, number of sessions, and session duration, as well as by comparator type (control/placebo vs. active treatment), with each moderator tested in a separate (univariable) mixed-effects model. The dependent variable was the standardized mean difference (d) for pain reduction, and categorical moderators were tested in separate univariable models (one moderator per model) (Table 4).

Table 4.

Results of the weighted ANOVAs for the influence of categorical variables on the effect sizes.

Moderator variable	k	d+	95 % CI		ANOVA results
			dl	du
Frequency					F(1,52) = 0.267, p = 0.607
Meets pre-specified criteria	45	0.159	−0.016	0.333	R2 = 0.00
Does not meet pre-specified criteria	9	0.266	−0.112	0.645	QE(52) = 159.640, p < 0.001
Intensity					F(1,42) = 1.553, p = 0.2196
Meets pre-specified criteria	30	0.254	0.082	0.427	R2 = 0.00
Does not meet pre-specified criteria	14	0.075	−0.158	0.309	QE(42) = 88.664, p < 0.001
Sessions duration					F(1,54) = 11.334, p = 0.001
Meets pre-specified criteria	21	0.465	0.236	0.694	R2 = 0.243
Does not meet pre-specified criteria	35	−0.013	−0.182	0.156	QE(54) = 146.526, p < 0.001
Number of sessions					F(1,51) = 0.001, p = 0.981
Meets pre-specified criteria	40	0.152	−0.039	0.343	R2 = 0.00
Does not meet pre-specified criteria	13	0.156	−0.158	0.471	QE(51) = 158.683, p < 0.001
Comparison groups for TENS					F(1,54) = 15.445, p < 0.001
Control or placebo	23	0.495	0.277	0.713	R2 = 0.235
Other active treatments	33	−0.041	−0.207	0.125	QE(54) = 148.202, p < 0.001

Note. ANOVA: analysis of variance; 95 % CI: 95 % confidence interval; k: number of comparisons between treatment and control group; d₊: pooled standardized mean change difference; dl: lower confidence limit for d₊; du: upper confidence limit for d₊; F: Knapp–Hartung statistic for testing the significance of the moderator variable; Q_E: statistic for testing the model misspecification; R²: proportion of variance accounted for by the moderator. Each categorical moderator was tested in a separate mixed-effects model (one moderator per model); estimates are not adjusted for other stimulation parameters.

Comparator-stratified analyses (secondary): session duration (p < 0.001) and comparator type (p < 0.001) were significant moderators of TENS efficacy. No significant moderating effects were found for frequency (p = 0.607), intensity (p = 0.220), or number of sessions (p = 0.981). Session duration explained 24.3 % of the variance: studies with sessions ≥40 min showed a substantial analgesic effect (d+ = 0.465, k = 21), compared to a negligible effect in studies with shorter sessions (d+ = –0.013, k = 35). Comparator type accounted for 23.5 % of the variance: TENS was significantly more effective than control or placebo (d+ = 0.495, k = 23). In contrast, no significant difference was observed when TENS was compared with other active treatments.

Across trials, only 6/36 met all four pre-specified dosing criteria concurrently, and 11/36 were not classifiable (NR/insufficient reporting) for at least one parameter; therefore, we did not perform a pooled composite ‘all-criteria met’ comparison due to limited power and potential selection bias. We then conducted an additional moderator analysis restricted to the subset of studies that administered TENS for more than 40 min per session (k = 21), to further explore dosage-related effects. Within this subset, only number of sessions showed sufficient variability for moderator analysis (n = 9 vs. n = 9).

A meta-regression within this subset found that the number of sessions was not significantly associated with effect size (F(1,16) = 1.031, p = 0.325, R² = 0.021). Residual heterogeneity remained significant (QE(16) = 37.480, p = 0.002), suggesting that other unexamined factors may contribute to variability in treatment outcomes. Other parameter-specific moderator analyses were not feasible due to limited variability, as most studies applied frequency and intensity settings classified as meets pre-specified criteria. Timing of pain assessment could not be examined as a moderator because it was insufficiently reported (only 11/36 trials assessed pain during/after a treatment session; only 2 during TENS; 9 post-session without reporting the elapsed time from stimulation end to measurement).

Egger's test indicated a statistically significant association between standard errors and corresponding effect sizes (t(54) = 2.456, p = 0.017), suggesting possible funnel plot asymmetry. The trim-and-fill method imputed 10 studies to correct for asymmetry, resulting in a reduced effect size that was no longer statistically significant (p = 0.803) (Supplementary Fig. S3). Given heterogeneity, funnel-plot asymmetry may reflect publication bias or other small-study effects [91]; therefore, the presence and magnitude of publication bias remain uncertain.

4. Discussion

The findings of this meta-analysis suggest that the clinical effectiveness of TENS for reducing KOA pain may depend on how it is applied. Session duration emerged as a key factor, with ≥40-min sessions associated with greater pain relief than shorter applications, whereas frequency, intensity, and number of sessions were not significantly associated with outcomes; limited variability and inconsistent reporting may have obscured their effects. Overall, these findings support the importance of treatment fidelity to pre-specified dosing criteria and suggest that suboptimal protocols, particularly insufficient session duration, may underestimate TENS analgesia in KOA.

Previous systematic reviews and meta-analyses of TENS for KOA have reported inconsistent findings, which may explain why several clinical guidelines recommend against its use [3,8,24]: two meta-analyses found no benefit [25,26], while two suggested potential analgesic effects [27,28]. None accounted for key stimulation parameters (frequency, intensity, session duration), limiting interpretability. In contrast, the present meta-analysis included 36 trials (2518 participants), more than doubling the sample of the most recent review [28], enabling a more robust evaluation of TENS efficacy and dosage. In the pooled analysis across all studies, TENS was not statistically superior to comparators; however, a pre-specified sensitivity excluding two influential trials yielded a statistically significant analgesic effect, while heterogeneity remained substantial. Accordingly, we examined comparator type and stimulation parameters as potential moderators. This is, to our knowledge, the first meta-analysis on TENS for KOA that integrates an overall efficacy estimate with a structured investigation of how stimulation parameters and comparator type influence outcomes.

Given the heterogeneity observed in the overall analysis, we examined the influence of the comparator type on TENS efficacy. We stratified analyses by comparator type to reduce heterogeneity introduced by active comparators (e.g., differing mechanisms, therapist interaction, or treatment intensity) and to better isolate the specific contribution of TENS. Notably, in trials comparing TENS with minimal or no intervention, the pooled effect size was statistically significant (p < 0.001). This effect was larger than the small effect observed in the full dataset after excluding two influential trials (p = 0.036), supporting a benefit of TENS relative to minimal intervention. In contrast, comparisons with other active treatments may underestimate the utility of TENS, as these comparators often involve co-interventions with established effectiveness. This does not imply inferiority; rather, any added value of TENS may be masked when comparators are themselves effective. In clinical practice, TENS is best positioned as an adjunct to core active care (education and exercise), providing short-term analgesia that may facilitate participation and adherence; whether TENS adds to exercise-based programmes remains uncertain.

Session duration emerged as a significant moderator of TENS efficacy, with studies applying stimulation for more than 40 min per session showing a markedly greater analgesic effect. This is consistent with a time-dependent response, with analgesia accumulating during stimulation and persisting thereafter, potentially reflecting endogenous opioid mechanisms [92]. Cheing et al. [37] found that 40 min of stimulation was sufficient to provide sustained KOA pain relief, whereas shorter sessions (e.g., 20 min) were less effective. In our dataset, extending sessions beyond 40 min did not clearly add benefit, suggesting 40 min as a pragmatic threshold, potentially best delivered via home use and/or immediately before and during exercise.

Frequency and intensity did not significantly moderate the analgesic effects of TENS in the overall analysis. This contrasts with evidence in fibromyalgia suggesting greater effects with high-frequency, strong-intensity stimulation [16]. These discrepancies could stem from differences in pain pathophysiology or from confounding by session duration in our sample. Notably, within the subset of studies that applied TENS for ≥40 min per session, most also used frequency and intensity settings that met the pre-specified criteria, leaving insufficient variability to evaluate their independent contribution. This overlap suggests session duration may be necessary for clinically meaningful analgesia, while other parameters may contribute in ways that could not be isolated here.

Regarding the number of sessions, no significant differences in effect size were found between studies delivering ≥10 sessions and those applying fewer treatments, even within the subgroup of trials using longer stimulation durations. These findings suggest that clinically relevant analgesia may occur early in KOA when stimulation parameters meet pre-specified criteria. This contrasts with prior evidence in fibromyalgia, where repeated sessions were found to be essential for achieving pain relief [16]. A possible explanation is that KOA pain is typically more localized, whereas fibromyalgia involves widespread pain and often central sensitization, which may require cumulative exposure to achieve comparable effects.

Two methodological aspects may have influenced our results and warrant further consideration. First, this review prioritized resting pain over movement-evoked pain when both were reported, as resting pain was the most consistently assessed outcome across studies. Although this improves comparability, it may undercapture effects on movement-evoked pain, which may be particularly responsive to TENS [34,36,93]. In addition, 25 of the 36 included studies did not assess pain at the expected peak of TENS analgesia, during or immediately after stimulation, which is the most appropriate window to evaluate peak TENS analgesia [34]. Moreover, timing was insufficiently reported to support a reliable moderator analysis: only 11/36 trials assessed pain during/after a treatment session, only 2 measured pain during TENS, and in the remaining 9 the post-session interval was not reported. Accordingly, outcome timing that does not align with peak efficacy may underestimate analgesia in functional scenarios [93,94]. These limitations highlight the need for future trials to assess movement-evoked pain and align outcome measurement with the expected temporal profile of TENS-induced analgesia.

Several limitations of this meta-analysis should be acknowledged. First, we focused exclusively on pain outcomes, as analgesia is the primary therapeutic goal of TENS. Other clinically relevant outcomes (e.g., function) were not synthesized; these may improve, at least in part, as a downstream consequence of pain reduction. Consistent with the primary symptomatic target of TENS, and to maximize outcome comparability across trials, we focused our quantitative synthesis on pain intensity. Second, our analyses suggest that the overall estimate of TENS efficacy may have been influenced by publication bias. Included trials were generally small (36 trials; 2518 participants; median total trial n = 55 [range 16–224] and median n per group = 20.5 [range 8–112]), which may amplify imprecision and small-study effects [36]. Egger's test and trim-and-fill suggested funnel plot asymmetry, but both are sensitive to the substantial heterogeneity in our dataset; thus, publication bias cannot be ruled out but should be interpreted cautiously. Nearly half of the included studies (47 %) were at high overall risk of bias, potentially compromising the reliability of the pooled findings. Accordingly, our pooled estimates should be interpreted in light of the high risk of bias in a substantial proportion of trials and the frequent incomplete reporting of key TENS delivery parameters, both of which can materially influence observed effects. RoB 2 concerns were driven primarily by D2 (deviations from intended interventions) and D4 (measurement of the outcome), largely reflecting incomplete blinding; only two studies implemented complete investigator blinding [49,50]. Future trials should prioritise feasible sham/active-placebo approaches and clearly reported blinding, particularly for outcome assessors. Moreover, the overall pooled effect was not statistically significant; a small effect emerged only after excluding two influential trials. Baseline pain was low (mean < 3/10) in six trials, raising potential floor effects that may have attenuated detectable improvements, however, the small number of such trials precluded a meaningful sensitivity analysis.

Although observed effects may be influenced by small-study signals and risk of bias, patterns observed in protocols meeting pre-specified dosing criteria highlight the importance of adequate delivery and reporting.

This meta-analysis suggests that TENS delivered for ≥40 min per session may provide short-term analgesia that supports engagement with active care. The absence of significant moderation by other parameters may reflect limited variability among protocols meeting pre-specified dosing criteria. Future trials should ensure adequate session length and intensity, assess pain during or immediately after stimulation, and include movement-evoked outcomes alongside resting pain. Further work should clarify other parameters under optimised conditions. Systematic reviews may benefit from pre-specified dosing criteria when interpreting evidence, as suboptimal application may underestimate the true efficacy of TENS, an approach previously emphasized in methodological discussions of TENS evidence synthesis [34,36].

Importantly, clinical guidelines should consider whether TENS delivery meets pre-specified, literature-informed treatment-fidelity criteria (and whether reporting is sufficient to assess this) when appraising TENS trials, while emphasizing the need for further well-designed adjunct trials that test incremental benefit over exercise and report stimulation parameters in sufficient detail, preferably including sham controls to quantify effects beyond contextual influences. Failure to account for dosing-related factors may lead to misleading conclusions and the potential exclusion of effective non-pharmacological options. By integrating findings from trials in which TENS delivery meets pre-specified, literature-informed treatment-fidelity criteria, guidelines can better fulfill their purpose of supporting evidence-informed decisions and improving patient outcomes.

Data availability statement

The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.

Declaration of Generative AI

During the preparation of this work, the authors used ChatGPT (OpenAI) to refine the academic English style and improve clarity, coherence, and readability of the manuscript text, including drafting alternative phrasings for revisions and responses to peer-review comments. The tool was not used to generate or analyze data, perform statistical analyses, or create figures or tables. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article.

Funding disclosure statement

This research was supported by grants from the University CEU Cardenal Herrera (INDI24/37 GIR24/29).

Competing interests

The authors declare no conflicts of interest.

Handling Editor: Professor H Madry

Appendix A. Supplementary data

The following is the Supplementary data to this article.