ANÁLISIS DE RIESGO RESIDUAL SEGURIDAD DE PARÁMETROS FÍSICO QUÍMICOS
4.3 AJUSTE ÓPTIMO DE LA INCERTIDUMBRE PREDICTIVA SEGÚN LOS CÓDIGOS ARÁBIGOS DE VARIABLES MULTIFUNCIONALES.
3.5.2.1 Multiple non-indq>aidait sites
Paiodontal q)idemiology must also consider many measuremaits of the same variable. These are site-based measures of aetiological factors (plaque, restorations or specific micro organisms) and disease outcome as well as subject based factors such as age, tobacco and health status. Consequently there has also been discussion about wh^her the unit of infCTence should be the site or the subject.
One prominent example of the use of site-based inference was by the Forsyth group in the development of the burst theories of attachment loss (Haffajee et al, 1983; Socransky et al,
1984; Haffajee et al, 1985). Simple statistical techniques assume independoit data, but the use of site-based units of inference ignores the mutual dependence of attachment loss at several sites in the same mouth on subject-based factors (Imrey, 1986). There is agreement that site-based inference detected many false positives and the statistical basis of the burst theories has been rq)udiated (Imrey, 1986; Ralls and Cohen, 1986; Gunsolley and Best, 1988; Sterne etal, 1990). Analytic techniques have been proposed which use the site as the unit of inference but consider the effect of subject based factors (Baelum et al, 1990; DeRouen etal, 1991; Albander and Goldstein, 1992), but these techniques are expaimental and complex.
Imrey (1986) and Sterne etal. (1990) have recommended the use of subject-based units of inference, particularly if a subject-based factor is undo* investigation (e.g. HTV infection). Subject-based units are suitable for parametric and non-parametric analysis including regression with or without transformed variables but there is loss of precision (Stome et al,
1990). For example, the use of intra-subject means swamps important detail if there are important changes at small numbers of sites. There is an additional problem that parametric
To prevent swamping, Imrey (1986) proposed sevrai methods for summarising important sites within an individual (i.e. summary measures) including:
i) the mean/medians of the k largest sites with attachment loss ii) the Ath pCTcentüe of site attachmoit level changes
iii) the proportion of sites with attachment loss above a given threshold iv) clinical categorisation of response
where k is small. It will be seen that: (i) measures the most severe site i i k = 1; (iii) is the extent of disease within a subject and (iv) becomes the presence ^ a b s e n c e of a characteristic for a binary outcome.
The final consequence of the number of factors and outcomes involved in periodontal research is the number of possible combinations of variables which increases type 1 error. Type one a ro r can be reduced by limiting the number of analyses to a small number of outcomes stated in advance.
3.53 Measurement
There are numerous measures and indices of periodontal health. Early indices a g g r^ te d signs of gingivitis and periodontitis. Recent measures have separated the two.
Periodontal indices are prone to poor rehability. The effects of poor measurement reliability can be reduced by the use of appropriate measures or indices and by calibration. The uncertainty created by poor rehabihty can be quantified during the study by comparing rq>eat examinations (RamÇord, 1967; Barnes etal, 1986; Pihlstrom, 1992)
3.5.3.1 Aggregate measures
Aggregate measures were used extaisively in early q)idemiology of periodontal diseases but are used less now. Burt (1991) has discussed the problems of such indices and in particular Russell’s periodontal index. His main concan was their biological validity, at the time gingivitis was believed to be the early stage of periodontitis and paiodontitis was assumed to be generalised throughout the mouth. We know now that gingivitis is common but periodontitis is much less so, that most gingivitis does not become periodontitis and that periodontitis can be site-specific. In addition, the inclusion of gingivitis on the same scale created mathematical distortion of the findings. Group means wCTe also calculated with Russell’s PI and this swamped the shapes of distributions.
Methodology
The Community Paiodontal Index of Treatment Need (CPITN) is a recent aggregate index whose biological validity has been challenged (Ainamo et al, 1982; Baelum et al, 1995). CPITN is not sufficiently precise for research (Burt, 1991) and is unsuitable for research with HTV since it ignores recession. Attachmait loss in the absence of pocketing is characteristic of HTV-associated diseases (Winkler and Murray, 1987).
The only current altm ative to aggregated measures are to measure probing dq)ths, attachment loss and gingivitis separately (Burt, 1991). Partial scoring (simplified indices) reduces the burden to researcher and subject. They may include selected teeth, fewer sites on individual teeth or sections of the mouth (Ramgord, 1967; Barnes et al, 1986). Partial scoring under-records attachment loss and plaque and gingivitis scores so researchers must decide whether to sacrifice sensitivity for expediency (Steme et al, 1990; Bentley and Disney, 1995).
3.5.3.2 Attachment levels and probing depths
Periodontal attachment level is commonly recorded relative to a fixed reference point, usually the cemento-enamd junction (CEJ), by passing a fine probe down to the junctional
epithelium (Pihlstrom, 1992). Measurement is made on visual scales or electronically. In longitudinal studies measuring change in attachment level, customised appliances (stents) are sometimes used to provide a referaice point in both the long and horizontal axes of the teeth. Probing is complicated if the free gingiva obscures the CEJ. As a consequence, the probing dq)th is measured first, thoi the relative positions of the CEJ and gingival crest (RamQord, 1967). This has the advantage of recording both probing dqjth and recession (Griffiths etal, 1988), but adds problems of rounding error (Imrey, 1986).
Periodontal probing is not entirely satisfactory as a measurement method (Burt, 1991). Probing does not record the true site of the junctional q>ithelium and its validity changes with differing disease, probing force and probe width (Armitage et al, 1977; Lang, 1991; Thai and Heaney, 1991; Pihlstrom, 1992). To provide precise data many sites must be probed and this is uncomfortable and time-consuming (Watts, 1989).
Probing attachment loss (PAL) is not a reliable measure. Approximately 90% of recordings can be replicated within 1 mm (Badasten et al, 1984). This is a 2 mm range for a disease which infrequently destroys 6 mm of attachment in a lifetime. Reproducibility of probing changes with examiner (Smith etal, 1970), tooth type (EspelandafaZ, 1991), probing force (Mullaly and Linden, 1994), position of probe (Watts, 1989), site of exam (Kingman et al,
and can be increased with calibration, constant force probes, stents, appropriate marking, the use of the mean value of repeated examinations and automated probes (Smith et al,
1970; Magnusson etal, 1988; Watts, 1989; Watts, 1989; Espeland etal, 1991; Mullaly and Linden, 1994).
3.5.3.3 Gingivitis
Indices of gingival health have been reviewed extensively (Barnes et al, 1986; Grant et al,
1988; Greene, 1990). Current indices relate to gingival colour and oedema (Suomi and Barbano, 1968), bleeding on probing (Saxer and Muhlemann, 1975) or both (Muhlemann and Mazor, 1958; Loe and Silness, 1963). ’Gingival indices’ which consider recession or pocketing are aggregate indices (see above) (Schour and Massler, 1958).
The Gingival Index of Loe and Silness (Loe and Silness, 1963) has been used extensively and successfully in surveys and experimental q)idemiology, has explicit and mutually exclusive categories, is precise, lends itself to calibration and incorporates the presence of bleeding on probing (Alexander etal, 1971; Griffiths etal, 1988; Greene, 1990).
The reliability of gingival indices has received less research attention than measures of attachmait loss. Griffiths et al. (1988) expressed concern about reliability but Smith et al.
(1970) found the subjective measures were as reliable as more objective measures. Reliability improves with training and calibration (Smith etal, 1970; Alexander etal, 1971). Gingival indices are less reliable under field conditions (Kingman et al, 1991) and bleeding on probing is less rq>roducible in deeper pockets (Watts, 1987).
3.5.3.4 Plaque
Plaque indices have been used to record the extent (Greene and Vermillion, 1960; Quiglg and Hein, 1962; Greene and Vermillion, 1964; RamQord, 1967; Turesky et al, 1970), thickness (Silness and Loe, 1964) or weight (Grant etal, 1988) of plaque on a tooth.
The plaque index of Silness and Loe is sensitive to small changes, can be transformed easily to record the presence or absence of plaque and is widely used in epidemiological research (Greene, 1990). This index has subjective criteria and should be used by experienced and trained examines (Mandel, 1974).
Diagnostic criteria: Development and reliability
Ch a p t e r 4 De v e l o p m e n t a n d r e l i a b h^it y o f d i a g n o s t i c CRITERTA FOR PERIODONTAL CHANGES
ASSOCIATED WITH f f lV INFECTION
4.1 Abstract
Objectives
i, Develop diagnostic criteria for HTV-associated paiodontal changes and assess their ability to improve inter-examiner reliability.
ii, Identify how a panel of experts make diagnoses of HTV-associated periodontal changes.
Design
Formal experimental with pre- and post-testing of subjects randomly assigned to study and control groups.
Method
Recording signs and symptoms seen on clinical slides
Participants
5 oral medicine specialists and 4 periodontists of a univCTsity stomatology department with clinical and/or research experience of HTV-associated periodontal diseases.
Outcome measures
Reliability of recognition of clinical signs, signs associated with diagnoses and reliability of diagnoses.
Results
Kappas for recognising clinical signs ranged between -0.04 and 0.75. Signs associated with diagnoses did not correspond with classical characterisation of diseases. Diagnostic reliability improved in the study group more than the control group (P = 0.011, Wilcoxon rank sum) and reached a higher level than the controls (P = 0.005).
Conclusions
The examiners showed only fair reliability in the recognition of clinical signs, made diagnoses intuitively and had only fair agreement on the diagnosis of periodontal diseases. The inter-examiner reliability of examiners trained and calibrated in the use of the criteria increased and was greater than among untrained examiners.