Methodological problems of a prospective, controlled clinical study on the 3-year performance of two glass-ionomer cements

A recent prospective, controlled clinical study by Fotiadou et al. was published in the September 2019 issue of the Quintessence International journal [1]. The main conclusion of this study is that the "high viscosity glass-ionomer cement Equia Fil restorative system showed moderate to high failure rates after 3 years". This conclusion is misleading, due to several problems with the applied study method:

(i) Lack of appropriate comparator: Equia Fil was not compared to any current gold standard, such as composite resin or amalgam restorations. Therefore, the concluding judgement of a "high failure rate" is arbitrary, since it is not clear against which alternative as comparator or benchmark the observed failure rate was measured? Perhaps any placed amalgam or composite resin fillings would have yielded the same failure rate in the selected patient cohort? The study did compare Equia Fil against Fuji IX Fast GP+Fuji Coat LC and, separately for each material, between baseline and 3-year results. However, the former comparison yielded no significantly different results in restoration failures, hence, warranting at most the conclusion of a minimal failure rate of Equia Fil in relation to its comparator. The latter comparison follows an uncontrolled study design from which no conclusions concerning the restorative material as causal factor for restoration failure can be made [2].

(ii) Insufficient sample size: The aim of the study was to test the null-hypothesis that "there were no significant differences in the performance of the two GICs in the 3-year observation time". Since both type of materials where high-viscosity glass-ionomers, each coated with a resin based material, any possible difference in actual failure rate may have been assumed to be of small percentage. However, the percentage of any detectable statistically significant difference in treatment effect between competing medical interventions is dependent on the study’s sample size [3]. This means that the smaller the difference to be detected the larger the sample size of the study needs to be. The use of a simple formula for sample size calculation reveals that in order to detect a statistically significant (p < 0.05) difference of 10% or more, at 80% statistical power, a minimum sample size of N = 200 per intervention group is needed. This means that if the true difference in effect size lies below 10%, any study with a sample size of N = 200 or below will yield large confidence intervals at non-significant levels (p < 0.05) and thus is unable to detect such existing differences. Against this background and without reporting any sample size calculation, Fotiadou et al. included a total sample size of N = 85 tooth restorations (43 teeth restored with Equia Fil and 42 restored with Fuji IX Fast GP+Fuji Coat LC) at baseline. The chosen sample size was justified on basis of the ADA Acceptance Program Guidelines for Resin-Based Composites [4]. These guidelines define a minimum of 50 tooth restorations per product at baseline and 40 at 18 months but without providing any justification, statistically or clinically, for such set standard. Hence, it can be assumed that the reported lack of statistically significant difference in this study is due to the low sample size, i.e.: due to the inability of this study to detect any significant differences between the compared restoration types and not due to any clinical merits of the compared interventions.

(iii) Lack of effective randomisation: The study claims to be randomized but reports that "If one filling was applied with material A, automatically for the next filling material B had to be used." This represents alternation not randomisation. Alternation does not protect against selection bias, because it enables the trialist to know the allocation group of the next tooth in advance after the group allocation of the previous tooth by sealed envelope was determined. Here indirect selection bias may occur, facilitated by a combination of investigator discretion to enrol subjects or deny enrolment [5]. Hence, the risk of selection bias, i.e.: that patient characteristics would have confounded the study results, is high. Such risk was heightened by the reported practice that some patients receive more and some fewer fillings, thus resulting in the risk of clustering of individual patient characteristics in one treatment group above the other [1].

(iv) High loss-to-follow up: A high loss-to-follow up after 3 years may further have confounded the true results. From the 85 placed restorations at baseline only 54, with a 36.5% restorations loss to follow up, were available. Inclusion of the lost restorations could have substantially changed the reported results. If one considers 5 and 4 failed restorations for 29 Equia Fil and 25 Fuji IX Fast GP+Fuji Coat LC after 3 years, respectively, then adding the 14 lost Equia Fil as potential failures and the 17 lost Fuji IX Fast GP+Fuji Coat LC restorations as restoration survivals (or vice versa) would clearly favour one type of restoration above the other and negate the lack of reported statistically significant difference in the Ratio of restoration failure risk (RR) between the groups:

(1) Scenario based on the reported data [1]: RR 1.08; 95% CI: 0.32-3.58, p = 0.90;

(2) Worse case scenario I: Assuming the 14 lost Equia Fil as failures and the 17 lost Fuji IX Fast GP+Fuji Coat LC restorations as restoration survivals: RR 4.64; 95% CI: 1.72-12.50, p = 0.002 (in favour of Fuji IX Fast GP+Fuji Coat LC above Equia Fil);

(3) Worse case scenario II: Assuming the 14 lost Equia Fil as restoration survivals and the 17 lost Fuji IX Fast GP+Fuji Coat LC restorations as failures: 0.23; 95% CI: 0.10-0.56, p = 0.001 (in favour of Equia Fil above Fuji IX Fast GP+Fuji Coat LC);

Regardless the likelihood of the computed worse case scenarios I and II, they illustrate the lack of robustness of the reported study results, in order to withstand the challenge of high attrition bias due to the reported 36.5% loss-to-follow up of placed restorations.

In summary, the conclusions made by Fotiadou et al. in terms of the 3-year performance of glass-ionomers in Class II cavities in permanent teeth are undermined by lack of appropriate comparator, insufficient sample size, lack of effective randomisation and high loss-to-follow up and, thus, its value for guiding daily dental practice needs to be considered with caution.

References:

[1] Fotiadou C, Frasheri I, Reymus M, Diegritz C, Kessler A, Manhart J, Hickel R, Klinke T, Heck K. A 3-year controlled randomized clinical study on the performance of two glass-ionomer cements in Class II cavities of permanent teeth. Quintessence Int. 2019;50(8):592-602.

[2] Mickenautsch S, Yengopal V. Reports of uncontrolled clinical trials for directly placed restorations in vital teeth. Braz Oral Res. 2017 Jul 3;31:e48.

[3] Pocock SJ. Clinical trials. A practical approach. John Wiley & Sons Ltd. 1983. pp. 124-5.

[4] American Dental Association. Council on Scientific Affairs. Acceptance Program Guidelines for Resin-based Composites for Posterior Restorations. Chicago: ADA, 2001.

[5] Berger VW, Bears JD. When can a clinical trial be called ‘randomized’? Vaccine 2003; 21: 468–472.