Mineral derivatives in alleviating oral mucositis during cancer therapy: a systematic review

Search strategy

Only randomized controlled trials (RCTs) were included in the review from search databases: MEDLINE via OVID, EMBASE, CENTRAL, CANCERLIT via PubMed, and CINAHL via EBSCO (2000 to 11 September 2014; see ‘ Appendix A ’ for search strategies) based on PRISMA (Preferred Reporting Items for Systematic reviews and Meta-Analyses) and Cochrane guidelines (Higgins, Green & Cochrane Collaboration, 2008). The time frame loosely coincides with recent chemo-radiotherapy combinations from the year 2000 onwards e.g., 5-FU, cisplatin and radiotherapy—neoadjuvant chemoradiation; ECF. Trace minerals regarded essential by the World Health Organization guidelines (Food and Agriculture Organization of the United Nations, World Health Organization , 2004) were combined using Boolean filtered MeSH terms and text words “random,” “control,” and “cancer.” Authors were contacted if full text articles or Supplemental Information were missing. A meta-search strategy was applied to the World Wide Web for articles (conference proceedings, clinical practice guidelines, clinical trials, and pre-prints) not indexed in databases. There was no language restriction on trial eligibility.

Study selection

The articles underwent a preliminary extraction using a template (Appendix B), and pre-set criteria on study inclusion and exclusion (Table 1). Mineral derivatives were administered as oral rinse, tablet, or intravenously. Trials using other interventions administered concurrently (e.g., vitamins) were excluded. All ages, cancer types, staging, and cancer therapy combinations were included. Significantly biased studies were excluded in the analysis based on a bias appraisal of the strengths and weaknesses in selection and protocol.

Quality appraisal

Studies underwent a bias appraisal in accordance with Cochrane Collaboration guidelines using templates in Revman version 5.2 (Review Manager, 2014). Possible risks (selection, performance, detection, attrition, selective reporting, and other bias) were rated high, low, or unclear by the author and collated into a summary and graph. ‘Selective reporting’ assessed omitted information, and ‘other bias’ for causal analysis errors.

Outcome measures

The primary outcomes compared severity in independent variables: peak incidence, duration, and onset of oral mucositis (OM) between mineral derivative and control groups (standard care, or placebo) based on severity measures defined by the OM grading systems. The secondary outcomes were the incidence of pain (visual analog scale), and use of analgesics. Serum levels, total parenteral nutrition, and adverse events were discussed.

Statistical analysis

Dichotomous data was abstracted using a 2 × 2 contingency table for three independent conditions (participants with OM or not; pain or not; analgesics use or not), and continuous data abstracted by mean days, standard deviation, and sample sizes. Studies with no data for an outcome (e.g., some studies did not record analgesics use) were excluded. Studies with missing data and only p-values were converted into standardized mean differences using the Practical Meta-analysis Effect Size Calculator(online) (Wilson, 2014), and missing standard deviations imputed with similar data (Appendix C) (Wolter, 2007). All data was then pooled by outcome: peak OM incidence, OM duration, time to OM onset, pain incidence, analgesic use, and arbitrarily assigned a 95% confidence interval (CI) and expressed as Hedges’ adjusted g—decided a priori in correcting small sample biases and standardizing diverse protocols. A random effects model was decided a priori in account of all effect sizes via a summary estimate of mean distributions, and in minimizing random sampling errors between small and large studies when combining within and between study variances. All outcomes were expressed using the Comprehensive Metaanalysis version 2.0 (Borenstein et al., 2005). Heterogeneity was assessed using Cochran’s Q statistic with I² statistic above 50% indicating significant heterogeneity. Publication bias was assessed with Funnel Plots and Egger’s regression model (two-tailed; p-value less than 0.05 indicating significant bias) (Rothstein, Sutton & Borenstein, 2005).

Decision analysis

High quality trials were mapped to a decision tree using TreeAge Pro 2014 version R2.2 (TreeAge Pro , 2014). The risk of developing peak OM data was imputed by sensitivity, specificity, and pre-test or post-test Bayesian probability (Gelman, 2004). In the first node, equal weight (0.5) was assigned to treatment and controls for each mineral derivative arm. Sensitivity and specificity was calculated using MedCalc (Charlie’s Clinical Calculators, 2014) in conjunction with statistical program R (R Core Team, 2014) for subsequent nodes (2 × 2 contingency tables; participants with peak OM or not), and terminal nodes assigned pre-test (control), or post-test (treatment) Bayesian probability which defined ‘risk’ as the pre- or posttest predictive value in developing peak OM (Petitti, 2000). The pre-test (prevalence) of developing OM was set high “0.8” based on high risk groups developing OM during cancer therapy (Sonis, 2009). The tree was then “rolled back” in estimate of probable risks in developing peak OM or not, and in deciding which treatment was the most effective.

Quality assessment: selection and measurement risks

Random sequence generation was reported in six out of 16 studies via a computerized method, and allocation concealment in seven studies. Randomized methods were unstated or unclear in the remaining studies. Ten studies explicitly described double-blind methods, one study mentioned “double-blind” but did not specify details (Papas et al., 2003) and one study described blinding as impracticable given the different appearances in treatment and control agents (Markiewicz et al., 2012). Participant completion rates were high in 11 studies that reported attrition rates. Drop out reasons included death (n = 11); in one study, one in ten died during treatment with no follow-up (Madan et al., 2008). Treatment refusal or study withdrawal (n = 11), opting for herbal and alternative treatments (n = 2), progressive illness (n = 1), and administration errors (n = 6), were reported in 4% of participants, and 2% died during the course of treatment in the 11 studies with reported attrition rates. In selective reporting, missing domains in 14 studies (85% of participants) may have significantly biased some outcomes. Other biases in estimate of causal pathways assumed a direct causal link between serum levels and treatment successes based on assay samples alone. Three studies used active control agents (soybean or fluoride) and possibly biased outcomes (Papas et al., 2003; Lin et al., 2010; Lin et al., 2006). One study transferred four participants to the treatment group, potentially overcounting outcomes in the exposed group (misclassification differential bias) (Lambrecht et al., 2013). One study transferred patient files to the sponsor who then monitored outcomes (conflict of interest) (Buntzelet al., 2010). Figures 2 and 3 summarizes these findings in Revman version 5.2 (Review Manager, 2014).

Summary of trial and patient characteristics

The 16 included studies recruited 1120 participants, female to male ratio 1:2 in adults aged 18 to 83 years, mean age 49 years, and one study with children four to 18 years (Raphael et al., 2014). 40% included populations from Asiatic regions (Madan et al., 2008; Lin et al., 2010; Lin et al., 2006; Sangthawan, Phungrassami & Sinkitjarurnchai, 2013; Watanabe et al., 2010), 31% Europe (Markiewicz et al., 2012; Lambrecht et al., 2013; Buntzelet al., 2010; Raphael et al., 2014; Vokurka et al., 2005; Ertekin et al., 2004), 21% the Middle-East (Jahangard-Rafsanjani et al., 2013; Mansouri et al., 2012; Arbabi-kalati et al., 2012; Mehdipour et al., 2011), and 8% Anglo-America (Papas et al., 2003). Baseline characteristics were matched in nine studies with malignant neoplasms (head and neck cancer; all stages), and seven studies for all types of cancers and leukemia (Papas et al., 2003; Markiewicz et al., 2012; Raphael et al., 2014; Jahangard-Rafsanjani et al., 2013; Arbabi-kalati et al., 2012; Mehdipour et al., 2011; Vokurka et al., 2005). Tumor localizations and neutrophil count were unremarkable across studies. Nine percent of individuals exclusively underwent chemotherapy (Mansouri et al., 2012; Mehdipour et al., 2011), 22% radiotherapy (Madan et al., 2008; Sangthawan, Phungrassami & Sinkitjarurnchai, 2013; Ertekin et al., 2004), 29% chemo-radiotherapy (Lin et al., 2010; Lin et al., 2006; Lambrecht et al., 2013; Buntzelet al., 2010; Watanabe et al., 2010), 12% hematopoietic stem cell transplantation (HSCT) with or without specifying ALL or AML (refer to Table 2) (Papas et al., 2003; Arbabi-kalati et al., 2012), and 28% HSCT plus high dose chemotherapy conditioning (HDC) regimens (Markiewicz et al., 2012; Raphael et al., 2014; Jahangard-Rafsanjani et al., 2013; Vokurka et al., 2005). Chemotherapy combinations of unspecified dosages were reported in two studies as cisplatin, carboplatin, 5-fluorouracil, monoclonal antibody against epidermal growth factor receptor (EGFR), doxorubicin, dacarbazine, gemcitabine, methotrexate (Lambrecht et al., 2013; Arbabi-kalati et al., 2012). Radiation maximum exposures averaged between 60 to 70 Gy in all eight studies among individuals undergoing radiotherapy. HDC conditioning busulfan (4 mg/kg/d for 4 days; total 16 mg) and cyclophosphamide (60 mg/kg/d for 2 days; total 120 mg) were specified in two studies (Markiewicz et al., 2012; Jahangard-Rafsanjani et al., 2013) and treosulfan (42 g/m²) plus fludarabine (150 mg/m²) were specified in one study (Markiewicz et al., 2012).

Eleven studies (n = 841; 75%) administered the intervention before the onset of oral mucositis and when commencing cancer therapy (36% first cycle) (Madan et al., 2008; Lambrecht et al., 2013; Raphael et al., 2014; Sangthawan, Phungrassami & Sinkitjarurnchai, 2013; Arbabi-kalati et al., 2012) and all studies had the intention to treat (adjuvant or prophylactic) oral complications. Two studies were part of an ongoing study and imposed varying degrees of heterogeneity when compared to other studies (Lin et al., 2010; Vokurka et al., 2005). One study for children began the intervention at the onset of OM (Raphael et al., 2014). The intervention calcium phosphate was administered as the oral rinse Caphosol^® (dibasic sodium phosphate 0.032%, monobasic sodium phosphate 0.009%, calcium chloride 0.052%, sodium chloride 0.569%, purified water 30 ml) in three studies (Papas et al., 2003; Lambrecht et al., 2013; Raphael et al., 2014) four to ten times daily. One study used a supersaturated calcium phosphate rinse (SCPR) mixed in equal volumes calcium and phosphate aqueous solutions (Markiewicz et al., 2012). Zinc derivatives were zinc sulfate oral syrup 5 mg per cubic centimeter, three times daily (Sangthawan, Phungrassami & Sinkitjarurnchai, 2013); zinc sulfate capsules 50 mg zinc, two to three times daily (Mansouri et al., 2012; Arbabi-kalati et al., 2012; Ertekin et al., 2004); zinc sulfate mouthwash 0.2% oral rinse, twice daily (Mehdipour et al., 2011); zinc capsules 25 mg, three times daily (Lin et al., 2010; Lin et al., 2006); Polaprezinc (Promac granules^® 15%), four times daily (Watanabe et al., 2010). Selenium derivatives were administered as followed: selenium tablets 0.2 mg, twice daily (Jahangard-Rafsanjani et al., 2013); sodium selenite oral fluid 0.3 to 0.5 g an hour before radiotherapy sessions (Buntzelet al., 2010). Iodine was administered in the form Povidone-iodine 1% (betadine 1 ml) oral rinse two to four times daily (Madan et al., 2008; Vokurka et al., 2005), and one study used povidone-iodine as part of the standard care (see Table 2) (Markiewicz et al., 2012). Overall, 63% of the intervention was administered as oral rinse with the option of swallowing, and 37% in tablet form. Mineral derivative types were zinc derivatives (n = 549), calcium phosphate (n = 227), povidone-iodine (n = 228), and selenium (n = 116). There were three primary control conditions: standard care alone (n = 308), placebo capsules (n = 411), and placebo solutions (n = 401). OM severity was graded using variable scales: NCT-CTCAE (rated 0–5) assessed OM as adverse events in four studies (n = 262) (Lambrecht et al., 2013; Raphael et al., 2014; Sangthawan, Phungrassami & Sinkitjarurnchai, 2013; Watanabe et al., 2010); World Health Organization toxicity scale rated 0–4 in six studies (n = 455) (Markiewicz et al., 2012; Madan et al., 2008; Jahangard-Rafsanjani et al., 2013; Mansouri et al., 2012; Arbabi-kalati et al., 2012; Vokurka et al., 2005); four studies used a radiation toxicity scale- RTOG or ARMSC rated 0–4 (n = 263) (Lin et al., 2010; Lin et al., 2006; Buntzelet al., 2010; Ertekin et al., 2004); one study measured OM lesions using a Spijkevet scale rated 0–4 (n = 45) (Mehdipour et al., 2011); and a dental scale-NIDCR rated 0–5 (n = 95) (Papas et al., 2003). Visual Analog Scales were used to measure pain ranging from 0–5, 0–10, or 0–100. All outcomes were measured in days.

Oral mucositis severity outcomes

Of sixteen studies, thirteen studies reported OM peak incidence across OM severity grading systems (Fig. 4A). Two studies did not experience peak OM and were therefore excluded (Markiewicz et al., 2012; Ertekin et al., 2004). Another study imposed significant heterogeneity when included in the analysis (from I² = 61% to I² = 87%), possibly from a confounder (fluoride as part of standard care), and was therefore excluded (Papas et al., 2003). Participants who undertook mineral derivatives were less likely to experience peak OM than those without treatment (g = −0.47, 95% CI −0.7 to −0.2, p = 0.0006, I² = 61%). Despite high heterogeneity, spurious conditions were expected in underpowered studies, hence all studies were kept in the analysis for observing a general effect trend. OM mean duration were reported in six studies (Fig. 4B). However, three were excluded since substitute data artefacts (see Appendix C) caused marked heterogeneity (from I² = 0% to I² = 97%, I² = 82%, I² = 45%) (Papas et al., 2003; Markiewicz et al., 2012; Raphael et al., 2014). OM mean durations did not significantly differ between mineral derivative and control groups (g = −0.2, 95% CI 0.1 to −0.5, p = 0.128, I² = 0%). Conversely, times to OM onset reported in five studies were significantly delayed in treated participants (g = −0.5, 95% CI −0.8 to −0.2, p = 0.0002, I² = 35%; Fig. 4C) (Madan et al., 2008; Lin et al., 2006; Jahangard-Rafsanjani et al., 2013; Mansouri et al., 2012; Vokurka et al., 2005).

Pain and analgesics use

Six studies reported incidences of pain using a visual analog scale (0–5 or 0–10) (Markiewicz et al., 2012; Lambrecht et al., 2013; Sangthawan, Phungrassami & Sinkitjarurnchai, 2013; Arbabi-kalati et al., 2012; Watanabe et al., 2010; Vokurka et al., 2005). One study caused marked heterogeneity (from I² = 68% to I² = 96%) possibly from a different visual analog scale (0–100), and a confounder (fluoride as part of standard care) and hence was excluded in the analysis (Papas et al., 2003). Experiencing pain was less likely in treated participants (g = −0.5, 95% CI −0.1 to −0.9, p = 0.01, I² = 68%; Fig. 4D). However, pains experienced during cancer therapies were unlikely estimates of OM pain in isolation—an explanation as to why analgesic use was no different across groups (g = −0.01, 95% CI 0.7 to −0.7, p = 0.977, I² = 79%), and significant heterogeneity apparent in both pain incidences and analgesic use (Fig. 4E). One study alone did not significantly alter the sensitivity in analgesics use or in the six pain studies. Thus it was assumed included studies did not record pain or analgesics use at uniform time points, or measured OM pain in isolation from other cancer therapy pains.

Serum mineral levels (inflammation and infection)

Serum samples were significantly noted in seven studies (Papas et al., 2003; Lin et al., 2010; Lin et al., 2006; Buntzelet al., 2010; Sangthawan, Phungrassami & Sinkitjarurnchai, 2013; Jahangard-Rafsanjani et al., 2013; Vokurka et al., 2005). Serum selenium levels were measured in all selenium studies, and cancer therapy toxicities were less in treatment than control groups (stomatitis-less toxicity treatment 36.4% versus 23.5% control (Wilson, 2014); treatment 118.21 versus 94.78 nmol/min/mL plasma Glu.Px-scavenger radical for alleviating toxicities) (Buntzelet al., 2010; Mansouri et al., 2012). In three studies, zinc serum levels were also higher in zinc treatment groups than control groups (63 to 142 ± 31 µg/dl versus 69 to 117 ± 25 µg/dl post treatment serum zinc level) (Lin et al., 2010; Lin et al., 2006; Sangthawan, Phungrassami & Sinkitjarurnchai, 2013). Pro-inflammatory cytokines TNFα and Interleukins (IL-1, IL-2 and IL-6) reportedly play a role in developing OM, and formed a basis for measuring serum levels in two studies (selenium; zinc) (Jahangard-Rafsanjani et al., 2013; Watanabe et al., 2010). One of those studies formulated the treatment ‘polaprezinc’ based on reports it inhibited cellular signaling of TNFα but the study did not record serum levels (Watanabe et al., 2010). Infection and oral bacteria flora were observed in two zinc studies by the same author (Lin et al., 2010; Lin et al., 2006), and were also mentioned in one calcium phosphate study using microbiology tests for detecting S. mutans and Lactobacillus. S. mutans (Papas et al., 2003). There were also three excluded studies (Appendix D, Table 2) that measured bacteria in saliva samples but were not included in the review because the causal pathway on developing OM and oral bacteria was unclear, and posed a high risk of bias (Ertekin et al., 2003; Nagy et al., 2000; Meca et al., 2009).

Total Parental Nutrition (TPN) use

TPN were recorded in two calcium phosphate studies. In one study, there were no significant differences in the treatment and control groups (parental feeding p = 0.7; tube feeding p = 0.08; age range 4–18 years) (Raphael et al., 2014). In the other study, less TPN were used by treatment than control groups in an adult population (p = 0.02; 19–57 years) (Markiewicz et al., 2012).

Adverse events

Three studies discussed dysphagia (difficulty swallowing) in conjunction with oral mucositis scoring systems (Markiewicz et al., 2012; Lambrecht et al., 2013; Buntzelet al., 2010). In two calcium phosphate studies, there were no significant differences in severity between treatment and control groups (Markiewicz et al., 2012; Lambrecht et al., 2013). However, the selenium group had a significantly lower mean severity score of 1.5 versus 2.2 in the control group (RTOG radiation scale 0–4) (Buntzelet al., 2010). The same study did not observe a significant difference in xerostomia (dry mouth) and ageusia (taste loss). In one zinc study, xerostomia (p = 0.007; grade ≥2 CTCAE) and ageusia (p = 0.006; grade ≥2 CTCAE adverse events scale 0–5) showed significant improvement in treatment than control groups (Watanabe et al., 2010). Xerostomia also improved in the zinc treatment groups (p = 0.005; WHO criteria 0–4 adverse events) (Arbabi-kalati et al., 2012). Nausea and vomiting recorded in one zinc study were mostly mild, and abdominal pain, diarrhea, fever, lethargy did not occur in treatment groups (Sangthawan, Phungrassami & Sinkitjarurnchai, 2013). Neutrophil count and neutropenia were reported in three calcium phosphate studies but were not significant (Papas et al., 2003; Markiewicz et al., 2012; Raphael et al., 2014).

Publication bias

Publication bias was assessed for all outcomes but evidence of bias was not significant. Funnel plots by and large appeared symmetric—assuming outliers (outside funnel) were small study effects from small sample sizes. Egger’s test was also not significant: Peak OM incidence p = 0.07 (Fig. 5); OM mean duration p = 0.5; time to OM onset p = 0.9; pain incidence p = 0.4; analgesics use p = 0.7.

Decision tree

One study was included for calcium phosphate based on available raw data and a low risk of bias (n = 58) (Lambrecht et al., 2013); another study for selenium (n = 74) (Jahangard-Rafsanjani et al., 2013) under the same assumptions; three studies for zinc (n = 218) (Sangthawan, Phungrassami & Sinkitjarurnchai, 2013; Mansouri et al., 2012; Watanabe et al., 2010); povidone-iodine was not included due to high bias. The decision analysis suggests the calcium phosphate treatment groups were the most at risk of developing OM at 78% (close to 80% pre-treatment rate) followed by zinc 73%, and selenium 68%. The probable outcome between each arm (rate of true positives, false positives, true negatives and false negatives) at present does not reflect a recognizable trend (Fig. 6).