Abstract
Objective
Bleaching ( alias whitening) is increasingly the chosen means of modifying the appearance of teeth for cosmetic reasons. Behbehani et al. (2012) acquired the largest known data set for the effects of bleaching over time (300 patients for ∼1 year), but the analysis then was rather superficial and did not exploit the full depth of information available in it.
Methods
Data collected by Behbehani et al. (2012) were re-analysed to explore in detail the variation in outcome between patients and over time.
Results
There is a distinct regression in shade scale value with time, taking something over 1 year to approach stability. There appears to be an intrinsic shade that cannot be altered, at around 1 unit (0.5M1 on the Vita 3D bleached shade guide). Some patients (11/300 at 1 year) appear to be worse-off after treatment, especially those with lighter starting shades, while many get no real benefit, with a change ≤2 units.
Significance
The effects of bleaching are often transient, but the reversion to a darker shade than the original may be due to the cumulative damage done to the organic matrix of enamel. The ethics of the treatment are questioned, as are over-the-counter products and whitening toothpastes. It might be appropriate to suggest to a patient that there is no point unless the shade is darker than some (Vita) shade value, say ∼1M2, but in any case informed consent is in need of improvement. All observations of the effects of bleaching ought to be made on timescales longer than 1 year to be meaningful.
1
Introduction
Tooth-whitening is an increasingly popular cosmetic process that broadly is considered desirable by the public and accepted by the dental profession as a safe and reasonable treatment . This is despite the acknowledged non-normal appearance of ‘white’ teeth and the “Hollywood” smile . There are, however, aspects that receive, in our view, insufficient attention, despite the presence in the literature of adequate documentation.
One factor of concern in the treatment is the occurrence of sensitivity . That is, the treatment has irritated the pulp ( sensu latu ). This can only have occurred because of the existence of a path of communication for the active material, i.e. , hydrogen peroxide, through the enamel. This may be for two reasons. The enamel is itself necessarily permeable . It is essentially a composite structure, with crystallites of HAp embedded in a hydrous protein matrix, albeit to the extent of about 95% by volume. Were it not permeable, other than superficial discoloration from ingested materials could not occur, and thus bleaching of such subsurface color could not occur; bleaching would be no more effective than toothbrushing with abrasive. Secondly, the enamel is frequently (and increasingly with age) cracked . This leaves a more rapidly permeable pathway deep into the substance, to at least as far as the amelo-dentinal junction . Of course, the pulp is in direct communication at this layer through the odontoblastic processes and via the dentinal tubules themselves for diffusible materials . Further evidence lies in the observation that fully-dried enamel is less translucent than fully-imbibed material because of the light scattering that occurs at air spaces, where the change of refractive index is marked.
The existence of this permeability is well understood in dentistry. It is known to provide a reservoir for the peroxide that can subsequently interfere with the polymerization of bonding agents .
Thus, were it a matter simply of the natural permeability of the enamel, then control to limit the extent of permeation is in principle straightforward. Indeed, this is implemented implicitly through the restriction of the concentration of peroxide in ‘overnight’ at-home treatment kits, and the duration of ‘in-office’ treatments at higher concentrations . Thus, for one product range, the treatments with carbamide peroxide are: 8–10 h/day with 10%, 4–6 h/day with 15%, 2–4 h/day with 20%, 30 min/day with 35% . (Curiously, assuming a first-order dependence on concentration, the exposure, concentration × time, is far greater at low concentration than at high, maximum ratio 5:1, i.e. , reciprocity is not observed.)
There is a very large literature on the damage done to various proteins intracellularly by peroxide. There are in addition a variety of indications that macroscopically protein degradation occurs with hydrogen peroxide, and even such resistant materials as feathers (keratin) and pigskin (collagen) .
It can also be noted that such degradation is temperature-dependent, as would be expected from simple chemistry, and that temperature rise is a necessary consequence of exposure to light during so-called ‘light-activated’ bleaching .
Behbehani et al. reported the results from one of the largest studies of the considerable literature in the field, involving some 300 patients over 1 year. We may note that the numbers involved in other studies have ranged from 7 to 87 . Some of the general difficulties obtaining in such studies to account for such low numbers might include: patient recruitment, the failure of patients to attend follow-up, patients’ adherence to instructions, the cost of such studies, as well as the time and facilities needed. In addition, the calibration of investigators and the repeatability of color readings pose problems . The present purpose was to re-analyze that data with a view to understanding the variation over time of the effects observed, an aspect that was not fully explored then.
2
Material and methods, results and analysis
The full protocol for the primary data collection has been described elsewhere . The raw data were examined in full and treated without reference to the prior approach or its conclusions.
The data, in units of steps on the Vita 3D bleached shade guide (ranging over 0–14 for 0M1–5M3), were first re-expressed as the frequencies of each combination of ‘before’ and ‘after’ values for each observation time, viz. as a matrix ( T _{0 }, T _{x }) for x = 1–4, where the observations were made at 1–7 days, and ∼12, 24 and 48 weeks, respectively. This was adequate for the analysis, but visually it was found easier to grasp by rescaling to ( T _{0 }, T _{x }− T _{0 }) to express the change in shade, which may be taken as the relative efficacy of the treatment ( Fig. 1 ). Thus, the zero line for the difference ‘after’ corresponds to no change. Values above this line ( i.e. , positive) record a darker shade than the original, while below ( i.e. , negative values) indicate a continuing lighter shade.
Linear regression, using the frequencies as weights, then yields the fitted trend line for the difference at T _{x }as a function of initial shade. This is plotted bold (red) in Fig. 1 for each x , and then broken (purple) in each of the other diagrams, to show their relative positions. The values of the slopes of these regression lines are plotted in Fig. 2 against observation time. To these data was fitted an exponential function: y = a + b [1 − exp(− cx )], that is, an asymptotic approach to a limiting value, a + b , as being a parsimonious representation of the shape of the trend line. The analysis is shown in Table 1 .
r ^{2 } | Fit std. err. | DF | F | P |
---|---|---|---|---|
0.999633 | 1.4247 × 10 ^{−3 } | 1 | 1361.2 | 0.017 |
Param. | Value | Std. error | t | 95% conf. limits | P | |
---|---|---|---|---|---|---|
a | −0.61132 | 1.51 × 10 ^{−3 } | −406.16 | −0.63044 | −0.59220 | 1.0 × 10 ^{−5 } |
b | 0.11061 | 2.66 × 10 ^{−3 } | 41.51 | 0.076751 | 0.14448 | 5.8 × 10 ^{−4 } |
c | 0.05291 | 3.25 × 10 ^{−3 } | 16.30 | 0.011676 | 0.09414 | 3.7 × 10 ^{−3 } |
However, treating such data in aggregate in this fashion loses information in the sense that there is a time series for each individual subject. Thus, sorting the list primarily on initial shade ( T _{0 }), then that at T _{1 }within T _{0 }values, then that at T _{2 }within T _{1 }values, etc ., allows a view of the shade trajectories overall, although this is hard to discern in the raw data. However, fitting an (arbitrary) average surface (Loess, sampling proportion 0.6, polynomial degree 2; SigmaPlot 13, Systat Software, San Jose, CA, USA) to these ordered data does provide a sense of the overall effect ( Fig. 3 ), with reversion of the order of 1–2 shade units by T _{4 }, not much affected by the value at T _{0 }. Again, this is an averaging, so taking the fitted regression line for each case, using log(observation time) to reduce the severity of the effect of the initial high rate of reversion that is apparent on inspection, we obtain the set of lines shown in Fig. 4 , with the distribution of slopes shown in the underlying histogram.
2
Material and methods, results and analysis
The full protocol for the primary data collection has been described elsewhere . The raw data were examined in full and treated without reference to the prior approach or its conclusions.
The data, in units of steps on the Vita 3D bleached shade guide (ranging over 0–14 for 0M1–5M3), were first re-expressed as the frequencies of each combination of ‘before’ and ‘after’ values for each observation time, viz. as a matrix ( T _{0 }, T _{x }) for x = 1–4, where the observations were made at 1–7 days, and ∼12, 24 and 48 weeks, respectively. This was adequate for the analysis, but visually it was found easier to grasp by rescaling to ( T _{0 }, T _{x }− T _{0 }) to express the change in shade, which may be taken as the relative efficacy of the treatment ( Fig. 1 ). Thus, the zero line for the difference ‘after’ corresponds to no change. Values above this line ( i.e. , positive) record a darker shade than the original, while below ( i.e. , negative values) indicate a continuing lighter shade.
Linear regression, using the frequencies as weights, then yields the fitted trend line for the difference at T _{x }as a function of initial shade. This is plotted bold (red) in Fig. 1 for each x , and then broken (purple) in each of the other diagrams, to show their relative positions. The values of the slopes of these regression lines are plotted in Fig. 2 against observation time. To these data was fitted an exponential function: y = a + b [1 − exp(− cx )], that is, an asymptotic approach to a limiting value, a + b , as being a parsimonious representation of the shape of the trend line. The analysis is shown in Table 1 .
r ^{2 } | Fit std. err. | DF | F | P |
---|---|---|---|---|
0.999633 | 1.4247 × 10 ^{−3 } | 1 | 1361.2 | 0.017 |
Param. | Value | Std. error | t | 95% conf. limits | P | |
---|---|---|---|---|---|---|
a | −0.61132 | 1.51 × 10 ^{−3 } | −406.16 | −0.63044 | −0.59220 | 1.0 × 10 ^{−5 } |
b | 0.11061 | 2.66 × 10 ^{−3 } | 41.51 | 0.076751 | 0.14448 | 5.8 × 10 ^{−4 } |
c | 0.05291 | 3.25 × 10 ^{−3 } | 16.30 | 0.011676 | 0.09414 | 3.7 × 10 ^{−3 } |