Mesenchymal stem cell therapy for radiation-induced xerostomia: a systematic review and network meta-analysis

Abstract

Background

Radiation-induced xerostomia (RIX) is a frequent, debilitating complication of head and neck radiotherapy for cancer. Preclinical studies suggest that mesenchymal stem cells (MSCs) may protect and regenerate salivary glands, but clinical evidence remains fragmented. This study evaluates the safety and efficacy of MSC therapy for RIX patients.

Methods

Comprehensive searches of PubMed, Wiley Online Library, Cochrane, and CNKI were conducted up to July 2025 to identify relevant clinical studies. Two investigators independently screened records. A total of seven trials (n = 360 participants) were included. Meta-analyses were conducted using RevMan 5.4 and R Studio, with unstimulated whole salivary flow rate (UWS) as the primary endpoint. Secondary endpoints included stimulated whole salivary flow rate (SWS), Xerostomia Questionnaire (XQ) scores, and serious adverse events (SAE). Meta-analyses were conducted using RevMan 5.4 and R 4.5.1, with UWS as the primary endpoint. Heterogeneity was assessed by I² and large-study effects by Egger’s test. The protocol was registered on PROSPERO (CRD420250521958).

Results

Pooled analysis of the seven trials showed a statistically significant but clinically negligible increase in UWS with MSCs compared to controls (WMD = 0.02 mL/min, 95% CI: 0.00 to 0.03, p = 0.04). No significant differences were found for SWS (WMD = – 0.12 mL/min, 95% CI – 0.28 to 0.04) or XQ scores (WMD = – 0.54, 95% CI – 1.96 to 0.88; p = 0.46). The risk of SAE was not significantly different between groups (OR = 1.96, 95% CI 1.00-3.84, p = 0.05). Substantial heterogeneity was observed (I² >90%). Exploratory network meta-analysis suggested that bone marrow-derived MSCs (BMMSC) might outperform adipose-derived MSCs (ADMSC), but this finding is hypothesis-generating due to being based on a single BMMSC study.

Conclusions

MSC transplantationresults in a statistically significant but clinically marginal improvement in UWS for RIX, with no significant increase in SAE. The current evidence does not support the superiority of MSC therapy over conventional management. Future large-scale trials are required to determine if optimized MSC strategies can achieve clinically meaningful benefits.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13287-025-04824-2.

Introduction

Head and neck cancer (HNC) is a global malignancy with rising incidence [1]. Its primary treatment often involves surgery. This is frequently combined with radiotherapy or chemotherapy. Despite advances in precision techniques, such as intensity-modulated radiotherapy, radiation-induced xerostomia (RIX) remains a debilitating complication. It affects 60–90% of survivors [2]. RIX results from irreversible damage to the salivary parenchyma. This leads to hyposalivation, dysphagia, oral infections, and a profound reduction in quality of life [3]. Conventional management of RIX focuses on symptom palliation and prevention. For example, sialagogues such as pilocarpine can temporarily improve salivary flow but have inconsistent efficacy and cholinergic side effects [4, 5]. Meanwhile, radioprotectors (e.g., amifostine) and gland-sparing radiotherapy may reduce symptom severity, but do not reverse established glandular atrophy [6]. Additionally, novel approaches like acupuncture and photobiomodulation lack evidence for promoting structural regeneration [7, 8]. Overall, these strategies fail to address the underlying glandular atrophy, highlighting the need for regenerative therapies.

Mesenchymal stem cells (MSCs) represent a promising regenerative strategy due to their multipotent differentiation, immunomodulatory properties, and paracrine signaling, which facilitate tissue repair and angiogenesis [9]. In preclinical models of irradiated salivary glands, MSCs mitigate apoptosis, reduce fibrosis, and restore acinar structure by secreting growth factors and anti-inflammatory cytokines [10]. These findings have accelerated clinical translation, with recent trials evaluating autologous or allogeneic MSCs derived from bone marrow or adipose tissue [11]. A first-in-human pilot study of interferon-gamma-stimulated bone marrow-derived MSCs (BMMSC) reported a favorable safety profile and a trend toward increased unstimulated saliva production in patients with chronic RIX, with no dose-limiting toxicities [12]. Similarly, a phase II randomized controlled trial (RCT) involving adipose-derived MSCs (ADMSC) demonstrated a 38% increase in UWS, although patient-reported outcomes did not differ significantly from those of placebo [13]. Further supporting this field, complementary efforts, such as the optimization of clinical-grade salivary gland organoids, underscore the feasibility of stem cell-based glandular regeneration [14].

Despite these advances, clinical evidence remains sparse and heterogeneous. The observed functional improvements in human trials are generally modest compared to the robust regenerative effects seen in preclinical models. Limited direct comparisons between MSCs sources (for example, bone marrow vs. adipose) and delivery methods further complicate the interpretation of existing evidence. Previous systematic reviews have analyzed the efficacy of MSCs in treating animal models [10]. However, to our knowledge, there is currently no a comprehensive quantitative synthesis that evaluats the efficacy and safety of MSCs for RIX in humans through both conventional and network meta-analysis. The aim of this study was therefore to perform a systematic review, meta-analysis, and network meta-analysis (NMA) to evaluate the efficacy and safety of MSC therapy for RIX, synthesizing data from clinical studies to inform optimal strategies and guide future research.

Finally, it is also important to contextualize regenerative approaches within the broader management paradigm for RIX. Beyond regenerative strategies, nutritional and supportive interventions have been investigated to alleviate oral complications following radiotherapy, showing potential benefits in improving patient comfort and quality of life [15]. Furthermore, radiotherapy-related sequelae extend beyond xerostomia, encompassing periodontal and mucosal complications that represent a substantial burden in this population [16]. This evidence underscores the multifactorial nature of oral morbidity in HNC patients and suggests the potential need for integrative management strategies that combine regenerative, nutritional, and preventive measures.

Methods and materials

Protocol

This meta-analysis was conducted and reported in accordance with the PRISMA-NMA guidelines [17]. The study protocol was developed and prospectively registered on PROSPERO (CRD420250521958) prior to data synthesis.

Search methods

Comprehensive, highly sensitive electronic searches were executed by two authors (LSS and TXD) across four databases: PubMed, Wiley Online Library, Cochrane Library, and the China National Knowledge Infrastructure (CNKI). A predefined search syntax was adapted for each database using a combination of keywords and MeSH/Emtree terms related to (1) mesenchymal stem cells and (2) radiation-induced xerostomia. The complete search strategies for each database are provided in Supplementary Table S1. The search was restricted to studies published in English or Chinese. Initial searches were conducted on 20 October 2024, with a final update performed on 27 June 2025, to identify all relevant studies evaluating MSC therapy for RIX.

Inclusion and exclusion criteria

Inclusion and exclusion criteria adhered to PICOS principles: (1) participants: humans with RIX following head and neck cancer treatment. (2) interventions: therapeutic - grade MSCs (any source/delivery route) administered post-diagnosis. Both autologous and allogeneic MSC preparations were eligible for inclusion. In contrast, acellular therapies derived from MSCs, such as the secretome or exosomes, were excluded from this analysis. Our focus was specifically on cell-based therapies. (3) comparisons: placebo or untreated controls. (4) outcomes: mandatory reporting of unstimulated salivary flow rate; secondary endpoints included SWS, XQ, and SAE. (5) study design: randomized/controlled clinical trials (any publication year; Chinese/English). (6) exclusion criteria included: (a) xerostomia from Sjögren’s syndrome or other non-radiation etiologies; (b) duplicate publications, incomplete data, or studies of poor quality; (c) reviews, case reports, conference abstracts, or preclinical studies; (d) non-controlled designs or insufficient outcome data.

Data collection and risk of bias assessment

Two independent reviewers (WWL and SJK) extracted data using electronic forms, capturing: authorship, publication year, country, study design, sample size, eligibility criteria, participant characteristics, interventions, and outcomes. Mean and standard deviation (SD) values were extracted per treatment arm. When unavailable, these were estimated from median, interquartile range, standard error, p-values, or 95% confidence intervals per Cochrane handbook guidelines [18]. Inter-rater agreement for study inclusion was assessed using Cohen’s kappa, which was 0.88, indicating excellent agreement. Similarly, the risk of bias assessment was conducted independently [19]. For the risk of bias assessment, agreement between reviewers was also high (κ = 0.85). The quality of non-randomized trials was evaluated using the Newcastle Ottawa Scale (NOS) [20–22]. A NOS score of ≥ 6 indicates high-quality research, while studies with NOS values of 1–3 and 4–5 are defined as low-quality and moderate quality research, respectively [20]. Any discrepancies in data extraction or risk of bias judgments were resolved through a structured consensus discussion between the two reviewers. If a consensus could not be reached, a third senior reviewer (TZL) was consulted for adjudication. Final judgments categorized bias risk as ‘low,’ ‘unclear,’ or ‘high’ per RoB2 protocols.

Outcome measures

The primary outcome was the between-group difference in UWS change from baseline, prioritizing the latest follow-up interval. Secondary outcomes included: change in SWS, longitudinal variation in XQ scores, delivery methodology (e.g., intraglandular, intravenous), per-injection MSCs dosage (×10⁶ cells), and safety profile assessed via SAE. We defined serious adverse events in the included studies as: (1) a change in submandibular gland structure; (2) recurrence of past cancer or a new primary cancer; (3) any other adverse event assessed as Grade 3 or higher on the Common Terminology Criteria for Adverse Events (CTCAE) v5.0; or (4) any treatment-related adverse event (e.g., infection, oral discomfort/pain lasting > 1 week) that fulfilled any of the previous criteria.

Statistical analysis

Conventional meta-analysis

Conventional meta-analysis was performed using RevMan 5.4 software to evaluate the efficacy and safety of MSC therapy for RIX. Subgroup analysis was stratified by MSCs sources (bone marrow/adipose tissue) and per-injection doses. Given the anticipated clinical and methodological diversity, we planned a priori to include both RCT and non-randomized controlled clinical studies (e.g., case-controlled studies) to provide a more comprehensive overview of the available clinical evidence. We acknowledge that this may introduce heterogeneity, and thus, a random-effects model was preferentially used to account for this variability. The impact of study design on the results was explored through sensitivity analysis. Dose subgroup analysis was added as a post hoc exploration with an explicit rationale in the supplement. Continuous outcomes were expressed as weighted mean difference (WMD) or standardized mean difference (SMD) with 95% confidence intervals (CI); non-significance was concluded if the 95% CI included zero.

Network meta-analysis

Network meta-analysis: Frequency-based NMA was conducted using Stata 18.0 and R 4.5.1 under a common heterogeneity variance assumption. The validity of an NMA relies on the underlying assumptions of transitivity and consistency. The transitivity assumption (i.e., that the included studies are sufficiently similar in terms of effect modifiers, such as patient populations, radiation dose, and outcome definitions, to allow for valid indirect comparisons) was evaluated by comparing the distribution of clinical and methodological characteristics across treatment comparisons. The consistency assumption (i.e., agreement between direct and indirect evidence) was evaluated globally using the design-by-treatment interaction model and locally using the node-splitting method [23, 24]. Global heterogeneity was assessed via τ² estimates from NMA models [23]. Statistical significance threshold was P < 0.05. Funnel plots and Egger’s regression (n >10) tested publication bias and small-study effects for primary outcomes.

Results

Description of the search

Initial electronic searches identified 1,097 potentially relevant studies. Through stage-wise screening, seven clinical trials (4 RCTs, 3 case-controlled studies) encompassing 360 participants were ultimately included [12, 13, 25–29]. Among these, six investigations utilized ADMSC and one employed BMMSC, with comparators including placebo and untreated controls. The study selection workflow is detailed in Fig. 1.

Fig. 1.

Process diagram for systematic review and meta-analysis reporting standards (PRISMA)

The characteristics of the included studies are summarized in Table 1 and Table S2. The seven trials (4 RCTs and 3 case-controlled studies) enrolled a total of 360 participants. Key study characteristics, including participant demographics (mean age, time since radiotherapy), baseline radiation dose, MSC source (adipose or bone marrow), cell dose per injection, delivery method, and follow-up duration, are detailed in Table 1. All interventions utilized intraglandular injection, with per-administration doses ranging from 2.8 × 10⁶ to 25 × 10⁶ cells. Five studies reported SAE related to MSC therapy. Table 2 shows the outcome measures included in the study.

Table 1.

Study characteristics

Authors (Year)	Age/sex (m/f)	Study design	Groups	Irradiation (GY)	Days from radiation to MSC treatment (years)	MSC type and concentration	Administration route	Statistical analysis	Functional outcome	Follow-up (Months)
Gronhoj (2018)[26]	60.4,19/11	Prospective, randomized, controlled trial.	1. IR, treated with placebo, n = 15. 2. IR, treated with MSC, n = 15.	20-70.9	NR	2.8 × 106 ASCs/cm3/mL (abdomen of each participant)	Intraglandular injection (submandibular glands)	NR	USW significantly increased in the ASC-arm, but not in the placebo-arm, compared to baseline.	4
Lynggaard (2022)[25]	NR	Prospective, randomized, controlled trial.	1. IR, treated with placebo, n = 15. 2. IR, treated with MSC, n = 15.	60–68	2	2.8 million ASCs/cm3, isotonic NaCl (0.9 mg/mL) and human albumin (HA) 1% to a volume of 1 mL	Intraglandular injection (submandibular glands)	Repeated measures linear mixed-effects model	Unstimulated SFR increased to 0.20 and 0.16 mL/min in the ASC and placebo group, respectively, yielding a 0.05 mL/min (95% CI: 0.00–0.10; P = 0.051) difference between groups.	24
Lynggaard (2022a)[28]	59.5,7/3	A single-center pilot clinical trial	1. No-IR, healthy control, n = 10. 2. IR, treated with MSC, n = 10.	66–68	2	25 million cryopreserved AT-MSCs were injected into each sub-mandibu- lar and 50 million AT-MSCs into each parotid gland (bdominal adipose tissue from three healthy females)	Intraglandular injection (submandibular glands)	Mixed effects model	UWS increased from 0.13 mL/min at baseline to 0.18 mL/min. SWS increased from 0.66 mL/min at baseline to 0.75 mL/min.	4
Blitzer (2023)[12]	71,5/1	A single-center pilot clinical trial	Self Case Control, n = 10	37.7–68.2	2	10 × 106 MSC (iliac crest)	Intraglandular injection (submandibular glands)	Mixed-measures analysis of variance	a trend of increased salivary production	3
Jakobsen (2024)[13]	61,88/32	Prospective, randomized, controlled trial.	1. IR, treated with placebo, n = 60. 2. IR, treated with MSC, n = 60.	42–47	2	0.5 mL of ASCs (25 × 106 cells per gland) (bdominal adipose tissue from three healthy females)	Intraglandular injection (submandibular glands)	Ancova models	ASC treatment: UWS increase of 0.04 compared with pretreatment baseline placebo treatment: not increase 0.01.	4
Jakobsen (2024a)[29]	NR	A single-center pilot clinical trial	Self Case Control, n = 10	NR	2	ASC to the parotid glands (50 million ASC in each) and the submandibular glands (25 million ASC in each)	Intraglandular injection (submandibular glands)	Multilevel repeated- Measures mixed- effects model	SWS increased significantly from an average of 0.66 mL/min at baseline to 0.86 mL/minute, corresponding to an increase of 0.20 mL/min, or approximately 30%.	36
Carlander (2025)[27]	61.4 ± 7.1,88/32	Prospective, randomized, controlled trial.	1. IR, treated with placebo, n = 60. 2. IR, treated with MSC, n = 60.	NR	2	0.5 mL of ASCs (25 × 106 cells per gland) (three healthy females)	Intraglandular injection (submandibular glands)	Multilevel mixed- effects model	Treatment with ASC did not increase UWS compared with placebo.	4

NR: Not reported. IR: Irradiation-induced; ASC/AT-MSC/MSC: Autologous mesenchymal stem cells; UWS: Unstimulated whole salivary flow rate; SWS: Stimulated whole salivary flow rate; SFR: Salivary flow rate; HA: Human albumin

Table 2.

Include the outcomes measured in the study

Authors (Year)	Whole saliva production (mL/min)
	Con			MSC			Con				MSC
	Baseline						Treatment
	UWS	SWS	XQ	UWS	SWS	XQ	UWS	SWS	SAE	XQ	UWS	SWS	SAE	XQ
Grønhøj (2018)[25]	0.16 ± 0.09	0.95 ± 0.09	NR	0.12 ± 0.09	0.94 ± 0.09	NR	0.16 ± 0.04	0.95 ± 0.31	0	NR	0.18 ± 0.02	0.16 ± 0.30	0	NR
Lynggaard (2022)[26]	0.14 ± 0.04	0.54 ± 0.28	50 ± 11.56	0.11 ± 0.03	0.86 ± 0.21	51.5 ± 7.5	0.16 ± 0.02	1.16 ± 0.09	5	45.1 ± 2.87	0.20 ± 0.02	1.13 ± 0.09	6	35 ± 0.79
Lynggaard (2022a)[28]	0.45 ± 0.20	2.44 ± 1.06	3.13 ± 5.63	0.13 ± 0.02	0.66 ± 0.11	53.5 ± 6.7	0.45 ± 0.20	2.44 ± 1.06	0	3.13 ± 5.63	0.18 ± 0.02	0.75 ± 0.1	1	30.9 ± 6.7
Blitzer(2023)[12]	NR	NR	NR	NR	NR	NR	0.13 ± 0.17	0.50 ± 0.40	0	NR	0.19 ± 0.21	0.58 ± 0.43	0	NR
Jakobsen (2024)[13]	0.13 ± 0.06	0.99 ± 0.60	47.9 ± 19.9	0.13 ± 0.05	1.14 ± 0.59	51.0 ± 20.5	0.01 ± 0.01	20.89 ± 10.03	3	5.12 ± 1.91	0.04 ± 0.01	37.92 ± 10.03	12	5.90 ± 1.90
Jakobsen (2024a)[29]	NR	NR	NR	NR	NR	NR	0.13 ± 0.02	0.66 ± 0.12	0	53.5 ± 6.6	0.15 ± 0.02	0.86 ± 0.12	2	32.5 ± 6.6
Carlander(2025)[27]	0.13 ± 0.06	0.99 ± 0.60	47.9 ± 19.9	0.13 ± 0.05	1.14 ± 0.59	51.0 ± 20.5	0.02 ± 0.01	0.08 ± 0.03	7	2.74 ± 1.17	0.02 ± 0.01	0.04 ± 0.03	6	3.12 ± 1.18

Con: Control. MSC: Mesenchymal stem cell. UWS: Unstimulated whole salivary flow rate. SWS: Stimulated whole salivary flow rate. XQ: Xerostomia Questionnaire. SAE: Serious adverse events. NR: Not reported

Risk of bias assessment

The methodological quality of the four included RCTs was assessed using the.

Cochrane RoB2 tool (Fig. 2), and the three non-randomized studies were evaluated using the Newcastle-Ottawa Scale (Table S3). In summary, for the RCTs, four studies were judged as having a “low” risk of bias (Fig. 2). For the non-randomized studies, all three achieved a NOS score of ≥ 6, indicating high methodological quality (Table S2).

Fig. 2.

Risk of RCT bias assessment diagram

Outcomes

Outcome of conventional meta-analysis

Meta-analysis revealed a statistically significant but clinically minor improvement in UWS following MSC therapy compared to controls (WMD = 0.02 mL/min, 95% CI 0.00 to 0.03, p = 0.04). In contrast, SWS showed no significant intergroup difference (WMD = – 0.12 mL/min, 95% CI: -0.28 to 0.04, p = 0.13) (Fig. 3A-B). Similarly, no significant therapeutic benefit was observed in patient-reported XQ scores (WMD = – 0.54, 95% CI – 1.96 to 0.88; p = 0.46) (Fig. 3C). The incidence of SAE was not significantly different between the MSC and control groups (OR = 1.96, 95% CI 1.00 to 3.84, p = 0.05) (Fig. 4A). Substantial heterogeneity was detected (I² = 94%-96%, random-effects model, p < 0.00001) (Fig. 3).

Fig. 3.

Forest plots for the efficacy of MSC therapy on salivary outcomes in patients with RIX. A Forest plot of UWS. B Forest plot of SWS. C Forest plot of XQ scores. P < 0.05. ACI: Confidence interval. MSC: Mesenchymal stem cell. SWS: Stimulated whole salivary flow rate. UWS: Unstimulated whole salivary flow rate. WMD/MD: Weighted mean difference. XQ: Xerostomia questionnaire

Fig. 4.

Safety and subgroup analysis of MSC therapy for RIX. A Forest plot comparing the incidence of SAE between the MSC therapy group and the control group. The pooled OR with 95% CI is shown. B Forest plot of subgroup analysis of the primary outcome, UWS, stratified by the source of MSCs (adipose-derived, ADMSC; bone marrow-derived, BMMSC). ADMSC: Adipose-derived MSC. BMMSC: Bone marrow-derived MSC. CI: confidence interval. OR: Odds ratio. SAE: Serious adverse events. UWS: Unstimulated whole salivary flow rate. WMD/MD: Weighted mean difference

Subgroup analysis stratified by MSCs source (adipose vs. bone marrow), per-injection dose (10 × 10⁶, 25 × 10⁶, 2.8 × 10⁶ cells), study type (RCT vs. case-control), and follow-up time (≤ 4 vs. >4 months) revealed no statistically significant associations with salivary restoration (p > 0.05). Extreme heterogeneity was observed specifically in the ADMSC and dose subgroup (25 × 10⁶ cells), with I² = 95–96% (p < 0.00001). This heterogeneity likely stems from limited BMMSC evidence (n = 1) confounding comparative interpretation (Fig. 4B and Fig. S1–3).

Outcome of NMA

The network diagram (Fig. S4) illustrates the evidence structure of the NMA comparing MSCs with placebo. Forest plots and the ranking table (Fig. 5 and Table S4) provide the estimates and corresponding 95%CI for different categories of MSCs compared with no treatment and placebo in the NMA. The estimated efficacy compared with placebo is shown in Fig. S2. As an exploratory analysis, the NMA was conducted to generate hypotheses regarding differential efficacy between MSC sources. The results indicated that, compared with the no-treated group (NoTx), ADMSC (WMD = – 0.27 mL/min, 95% CI – 0.40 to – 0.14) and placebo (WMD = – 0.29 mL/min, 95% CI – 0.42 to – 0.16) were associated with reduced UWS. BMMSC showed a non-significant trend toward saliva recovery compared to NoTx (WMD = 0.06 mL/min, 95% CI – 0.11 to 0.23). Compared with placebo, BMMSC was associated with a significant increase in UWS (WMD = 0.35 mL/min, 95% CI 0.13 to 0.57), while ADMSC did not show a significant effect (WMD = 0.02 mL/min, 95% CI 0.00 to 0.04).

Fig. 5.

The NMA for UWS recovery. A, B Forest plots from the NMA showing WMD and 95% CIs for comparisons between different interventions. Interventions include ADMSC, BMMSC, placebo, and NoTx. C, D Surface under the cumulative ranking curve plots for all interventions. A higher cumulative ranking curve value (or P-score) indicates a higher ranking for efficacy in improving UWS. ADMS: Adipose-derived MSC. BMMSC: bone marrow-derived MSC. CI: Confidence interval. NoTx: No-treated group (NoTx). NMA: Network meta analysis. UWS: Unstimulated whole salivary flow rate. WMD/MD: Weighted mean difference. Note: The estimate for BM-MSC is based on a single study arm within the evidence network. Consequently, its high ranking should be interpreted as exploratory and hypothesis-generating, not as conclusive evidence of superiority

In Fig. S4 (upper triangle), we present the results of the random-effects pairwise meta-analysis for different MSCs, placebo, and NoTx regarding UWS. In terms of UWS, ADMSC were less effective than BMMSC (WMD = -0.33 mL/min, 95% CI – 0.55 to – 0.11). BMMSC was more effective than placebo (WMD = 0.35 mL/min, 95% CI 0.13 to 0.57). Conversely, ADMSC significantly outperformed placebo (WMD = 0.02 mL/min, 95% CI 0.00 to 0.04) but not NoTx (WMD = – 0.27 mL/min, 95% CI – 0.40 to – 0.14).

The cumulative ranking probabilities numerically favored BMMSC (P-score = 0.083) for UWS recovery (Fig. 5C, D). However, it is imperative to interpret this result with extreme caution. This finding is strictly exploratory and hypothesis-generating, as the estimate for BMMSC is based on a single study arm within a sparse evidence network. The high P-score should be viewed as a mathematical projection of limited data rather than as robust comparative evidence.

The global inconsistency model did not detect any significant global inconsistency for any outcome (p > 0.05). The node-splitting model also did not reveal any significant local inconsistencies (p > 0.05, Table S6). The consistency test results are presented in Table S4, with a significance level set at p < 0.004. The funnel plot analysis shows symmetry and discreteness, and the results are stable without significant publication bias (Fig. S4).

Discussion

This systematic review and NMA demonstrated MSCs transplantation yields a statistically significant increase in UWS (WMD = 0.02 mL/min, 95% CI:0.00 to 0.03, p = 0.04) in patients with RIX, a finding which is consistent with the meta-analysis by Arisha et al. [30]. The marginal improvement in UWS suggests MSCs may primarily act through paracrine-mediated tissue protection rather than de novo regeneration, a hypothesis that aligns with recent organoid studies [14].

However, it is crucial to balance statistical significance with clinical relevance. The mean improvement of 0.02 mL/min falls substantially below the proposed minimal clinically important difference (MCID) threshold of ≥ 0.1 mL/min for patients with RIX [31]. This threshold represents the smallest change perceived as beneficial by patients; the observed effect is unlikely to translate into meaningful symptomatic relief or improved quality of life and must therefore be considered clinically marginal. Regarding safety, the available evidence suggests a comparable risk of SAE between the MSCs and control groups, a finding supported by Grønhøj et al. [26] and Jakobsen et al. [29]. Consequently, current evidence does not substantiate MSC therapy as superior to conventional symptomatic management. The limited clinical impact observed suggests that the paracrine mechanisms activated by MSCs at the administered doses may be insufficient for functional restoration within the chronic, fibrotic microenvironment of irradiated glands. Future clinical development should therefore focus on strategies designed to exceed the MCID, such as repeated injections, higher cell doses, or combination regimens that enhance MSCs survival and paracrine activity.

Despite the limited clinical significance, our findings are consistent with preclinical evidence [32]. A prior meta-analysis in rodent models has reported similar improvements in salivary function. These benefits are attributed to MSCs, which promote acinar cell regeneration and reduce fibrosis mainly via paracrine signaling [33]. In human trials, ADMSC have been shown to consistently improve UWS by 33% at 1 month post-injection, as observed in the MESRIX series, which corroborates our pooled estimates [26]. BMMSC exhibited comparable efficacy, though with slightly higher variability in outcomes, potentially attributable to differences in cell potency and delivery routes [12]. Subgroup analysis stratified by MSCs source revealed no significant intergroup differences (p = 0.32). This suggests broad applicability across autologous and allogeneic preparations within the limitations of the current evidence.

The modest functional improvement, despite promising preclinical data, likely reflects significant pathophysiological challenges within the chronic radiation-induced microenvironment. Post-radiotherapy, salivary glands undergo extensive fibrotic remodeling, characterized by dense collagen deposition, vascular damage, and a pro-inflammatory state [34, 35]. Critically, this fibrotic niche, with its altered mechanical properties (e.g., increased extracellular matrix stiffness), may not only serve as a physical barrier but also actively dysregulate transplanted MSCs function through sustained activation of mechanosensitive signaling pathways [36]. The ensuing hypoxic and nutrient-deprived conditions could profoundly limit MSCs paracrine activity, while the physical barrier of scar tissue may prevent crucial cell-cell communication and the effective distribution of trophic factors to remaining viable parenchyma [37]. This supports the concept that MSCs efficacy is highest in acute or subacute injury models where the microenvironment is more permissive. Moreover, MSCs secrete growth factors like vascular endothelial growth factor and hepatocyte growth factor, which promote angiogenesis and epithelial proliferation [38]. Exosome-based therapies further amplify these benefits by delivering microRNAs that inhibit pro-fibrotic pathways, leading to improved gland architecture histologically [39]. Dose-dependent responses were apparent in our meta-regression, with lower per-injection doses (2.8 × 10⁶ cells/cm³) yielding greater UWS gains (WMD = 0.03 mL/min, 95% CI: 0.02 to 0.05, p < 0.001). Therefore, the clinical marginality of our results may not inherently reflect a failure of MSC therapy but rather highlight the imperative to develop strategies that can overcome these hostile conditions, such as co-administration with anti-fibrotic agents or using pre-conditioned MSCs tailored for survival in fibrotic niches.

Heterogeneity in our meta-analysis was notably high (I² >90% for UWS). This substantial variability can be attributed to several clinical and methodological factors, including differences in MSCs source (adipose vs. bone marrow), cell dosing regimens, manufacturing protocols, and delivery methods. Furthermore, considerable clinical heterogeneity existed across studies regarding patient factors (e.g., baseline radiation dose to the salivary glands, time since radiotherapy completion defining chronicity of injury, and baseline salivary function), and follow-up duration. The chronic, fibrotic microenvironment of irradiated salivary glands itself, characterized by altered extracellular matrix stiffness and a pro-inflammatory state, may also be a significant, yet unmeasured, source of heterogeneity affecting MSCs’ efficacy differently across patients. These methodological strengths enhance the reliability of our conclusions compared to earlier reviews that lacked subgroup stratification. Moreover, authors of included studies [12, 25, 27] hold patents related to MSCs technologies; this may introduce intellectual bias.

Our analysis revealed a comparable incidence of SAE between MSCs and control groups, supporting the short-term safety of local MSCs administration. This finding contrasts with some preclinical concerns about MSC-induced cell proliferation [10], a discrepancy that may relate to the context of late radiation sequelae. Subgroup analysis detected no significant efficacy differences between BMMSC and ADMSC (p = 0.62), attributable to limited direct comparative evidence and underpowered BMMSC data (n = 1 study). While MSC therapy collectively demonstrated clinical benefits for RIX, source-dependent efficacy differentials remain unestablished. Therefore, we discuss which MSC-based therapeutic strategy is effective for RIX. Due to the limited number of studies, we conducted an NMA only for BMMSC and ADMSC. In the indirect comparison, UWS was numerically higher for BMMSC than for ADMSC, but the heterogeneity of both outcomes exceeded 90% significantly (95% and 96%, respectively). Given the exploratory nature of NMA, its results should be interpreted with caution. The NMA, while methodologically executed, was fundamentally limited by an imbalanced and sparse evidence network. The results for BMMSC, based on a single study, should be considered strictly exploratory and hypothesis-generating. They indicate a signal that justifies dedicated, large-scale, direct-comparison RCTs between MSCs sources, but do not provide sufficient evidence to recommend one source over another. The high P-score for BMMSC is a mathematical artifact of its point estimate in a fragile network and should not be misinterpreted as clinical evidence [12]. The resulting P-score is a mathematical projection of this limited data rather than robust comparative evidence. Therefore, our analysis cannot establish the superiority of one MSC source over another. The value of this NMA lies not in providing a definitive answer, but in highlighting a potential differential effect that justifies and should guide the design of future, large-scale, direct-comparison randomized controlled trials. Any biological plausibility for differences (e.g., enhanced immunomodulatory capacity [40] and differentiation capacities [41]) remains entirely theoretical in the context of RIX until such direct evidence is available.

It is essential to contextualize MSC therapy within the rapidly evolving field of regenerative medicine for RIX. Notably, salivary gland organoid transplantation has emerged as a promising alternative. Jeon et al. [14] recently demonstrated that transplantation of bioengineered organoids derived from resident stem cells could achieve a UWS improvement of 0.15 mL/min in a preclinical model [35]. This improvement approaches clinical relevance and exceeds the effects observed in our meta-analysis. Although direct comparisons remain limited by the lack of human organoid data, this approach represents a distinct regenerative strategy. While MSCs primarily modulate the microenvironment through paracrine signaling, organoids aim to directly repopulate the gland with functional, organized epithelial structures. The choice between these strategies may eventually depend on the extent of glandular damage. MSC-based immunomodulation may suffice in cases of moderate injury, whereas severe fibrosis might require the structural replacement offered by organoids. Future research should focus on direct comparative effectiveness and potentially even combinatorial approaches.

While MSC therapy offers a biologically plausible avenue for salivary gland regeneration, its role should be considered within a wider spectrum of adjunctive strategies that address the complex oral side effects of radiotherapy. Recent clinical data from nutritional interventions and oral health studies illustrate that supportive care can modulate treatment-related morbidity and improve quality of life [15, 16]. Consequently, MSC transplantation may ultimately serve best as part of an integrative management plan that combines regenerative, nutritional, and preventive measures.

Despite these advancements, limitations must be acknowledged. First, the small number of included trials (n = 7) and the sparse, imbalanced evidence network, particularly for BM-MSC (based on a single study), limit the robustness and generalizability of our conclusions, especially those from the NMA. Second, the high statistical heterogeneity (I² >90%), despite the use of random-effects models, suggests that the pooled estimate should be interpreted with caution. Third, the relatively short follow-up duration of most studies precludes a robust evaluation of long-term safety (e.g., risk of secondary neoplasms) and sustained efficacy. Fourth, methodological concerns regarding inadequate allocation concealment and blinding in some studies may have introduced performance and detection bias, particularly for patient-reported outcomes (XQ). Finally, the heterogeneity in MSC sourcing, manufacturing, and dosing, alongside variations in patient populations, constitutes potential confounding that our analysis could not fully adjust for. To address these, we propose: (1) multicenter, large-scale randomized controlled trials with internationally recognized and rigorous designs; (2) quantitative sialometry integrated with MR sialography and salivary proteomics for longitudinal efficacy monitoring; (3) Three-arm trials comparing viable cells vs. acellular derivatives (exosomes/vesicles) to delineate therapeutic advantage boundaries. This suggests that the transition from promising preclinical models to the clinic is hampered by the critical difference in disease chronicity and tissue microenvironment, highlighting a major challenge in the field of regenerative medicine.

Conclusions

In conclusion, MSC transplantation demonstrates statistically significant but clinically marginal improvement in unstimulated salivary flow without excess safety risk. Current evidence does not support superiority over conventional management, and further large-scale multicenter RCTs are required to establish clinical relevance.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Conceptualization: Shen-sui Li. Methodology: Xu-dong Tian and Quan-zhou Liu. Software: Shen-sui Li and Xu-dong Tian. Validation: Shen-sui Li and Ju-kun Song. Formal analysis: Wei-li Wang. Investigation: Ju-kun Song. Resources: Shen-sui Li. Data curation: Xu-dong Tian. Writing – original draft preparation: Shen-sui Li. Writing – review and editing: Xu-dong Tian and Wei-li Wang. Visualization: Ya-dong Wu. Supervision: Shen-sui Li. Project administration: Zheng-long Tang. Funding acquisition: Shen-sui Li. All authors have read and agreed to the published version of the manuscript.

Funding

This work was supported by the Natural Science Foundation of Guizhou Province (ZK [2025] 519); the Science-Technology Fund of Health Commission Guizhou Province (gzwkj2024-195); the Transverse project fund of the affiliated stomatological hospital of Guizhou Medical University (GYKQ2022HXZD05, GYKQKY202305); Teaching Reform Program of Higher Education Institutions of Guizhou Provincial Department of Education (2023143) and Key Project of Undergraduate Teaching Reform Research at Guizhou Medical University (JG2023017).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Competing interests

The authors declare no competing interests.