Correlation of nasal symptoms with objective findings and surgical outcome measurement

Thesis submitted for the degree of Master of Surgery, University of London, 1993.
Published (excluding Chapter 9) 1996.
Recompiled HTML format June 2007
© 1993 – 2014 JW Fairley

Mr James W Fairley BSc MBBS FRCS MS
Consultant ENT Surgeon

Chapter 5
The relationship between pain projected on a diagram of the face and systematically documented findings using rigid nasendoscopy

Chapter 5 Contents

Thesis Contents

Summary

The idea that intranasal abnormalities can cause facial pain and headaches, even in the absence of sinus infection, was proposed by Sluder (1918) and re-emphasised recently by Stammberger and Wolf (1988).

The concept has not gained universal acceptance.

Many practitioners remain sceptical about a nasal origin for headaches without documented evidence of sinus infection.

This study provides detailed statistical information on the relationship between pain at given sites and the findings on rigid nasendoscopy.

161 patients attending a nasal research clinic were analyzed. Using crosstabulation techniques and hierarchical log-linear modelling, significant associations were found between frontal headache and endoscopic abnormalities in the upper part of the nose, including the lateral aspect of the middle turbinate and the ostiomeatal complex (p <0.001).

The pain was located either centrally or on the same side as the abnormality.

Orbital pain and pain between the eyes was similarly associated with endoscopic abnormalities in the upper lateral part of the nose, and with septal spurs.

This study failed to demonstrate any association between high septal deviations, with middle turbinate contact, and facial pain or headache.


  • Mucosal pressure contact zones lateral to the middle turbinate were associated with frontal and orbital pain.
  • The presence of pus in the middle meatus was associated with pain over the maxillary sinus on the same side.
  • Polyps were not associated with pain at any site.

Although statistically significant relationships were demonstrated, their positive predictive value is low. Knowledge of the presence or absence of endoscopic abnormalities improved the accuracy of predicting the presence or absence of pain by an average of 6% in those sites where statistically significant associations were demonstrated. Many patients have pain without a corresponding abnormality, and many endoscopic abnormalities are found without a corresponding pain.

When a decision has to be made to treat a patient for facial pain and/or headache by correcting intranasal abnormalities, the weakness and variability of these relationships should be taken into account.


Introduction

It has long been recognized that abnormalities in the nose, particularly involving contact between apposing mucosal surfaces of the middle turbinate and its relations, can be associated with pains in the face, eye and head, even in the absence of sinus infection (Ewing and Sluder 1900; Yankauer 1908; Sluder 1927; Brown Kelly 1943,1946; Ryan and Ryan 1979; Stammberger and Wolf 1988). The concept of “Sluder headaches” has not gained universal acceptance, and many practitioners remain sceptical about a nasal origin for headaches without documented evidence of sinus infection (Couch 1988; Friedman and Rosenblum 1989). Others have gone to the opposite extreme, and believe that even tension headaches and migraines are caused by nasal mucosal congestion and can be treated surgically (Bonaccorsi 1988,1992; Hoover 1987,1992; Novak et al 1988; Novak 1992; Blondiau 1992). The latter group hypothesise that stenoses and mucosal contact zones in the roof of the nose and ethmoids predispose to venous stasis. This is said to result in a local accumulation of vasoactive chemical mediators such as histamine, serotonin and Substance P. Via anastomoses between the nasal and intracranial circulation, and also via trigeminal nerve axon reflexes, these mediators trigger migraine attacks. All report excellent results (80 to 100% success rates) for relief of headache by surgical or medical treatment of the nose, though none have carried out controlled trials.

My own experimental work with the nasal pressure probe (Fairley et al, 1992; Chapter 3 of this thesis) and the earlier work of Harold Wolff and colleagues (Ray and Wolff, 1940; Wolff, 1943) shows that the nasal mucosa is sensitive to pressure stimulation, and that the middle turbinate is more sensitive than other areas, but this was in healthy volunteers.

The purpose of this study is to provide detailed statistical information, in patients attending a nasal research clinic, on the relationship between pain at given sites and the findings on rigid nasendoscopy, and to test the hypothesis that endoscopic abnormalities at specific sites are associated with pains and headaches at corresponding sites.


Methods

Patients attending the author’s Nasal Research Clinic with a variety of nasal complaints (see Data appendices 1 and 5 for details) were asked to colour on a diagram of the head (figure 5.1), the sites and severity of pain. All patients attending the clinic were included, whether or not pain was a symptom. They were instructed to ignore the grid markings on the diagram.

Systematic rigid nasendoscopy (Stammberger, 1986) was then carried out by the author using a combination of Wolf 25 degree wide angle and 70 degree telescopes, 4mm and 2.7mm diameter, following topical anaesthesia and decongestion with 10% cocaine spray. The findings were recorded on a specially designed chart (figure 5.2).

The charts of the first 167 patients attending the clinic were analyzed. The sites of pain were simplified to 12 areas on a 3 X 4 grid and computer coded. The grid was made up of 3 vertical strips, Right, Left and Central, divided by two vertical lines passing through the medial canthi, and 4 horizontal strips, Frontal, Eye, Malar and Lower, divided by three lines passing through the upper orbital rim, the lower orbital rim and the angles of the mouth.

Endoscopic abnormalities were recorded diagrammatically on the chart in full detail, but for the purposes of computer coding and analysis endoscopic sites were simplified to 6 areas on 3 X 2 grid. The endoscopic grid was made of 3 vertical strips, Right, Central and Left, divided by vertical lines passing through the middle of the middle turbinates, and two horizontal strips, upper and lower, divided by a line passing just below the middle turbinate.

Statistical methods

For a “first pass” through the data, the type of abnormality was ignored and all endoscopic sites were coded as either 1 (abnormality present) or 0 (no abnormality). Pain sites were coded similarly, 1 for pain present, 0 for no pain.

Crosstabulations between all 12 pain sites and 6 endoscopic sites were calculated with Chi-squared tests, using the SPSS-PC computer program version 3.1, and the Chi-squared values with significance levels tabulated. A hierarchical log-linear model was then calculated for each pain site, using all the endoscopic information, to determine not only the effects of the endoscopic abnormalities individually but also their interactions. A backward elimination model was specified, with endoscopic sites being progressively discarded from the model until the reduction in Chi-squared was no longer significant at the 0.05 level.

After the first pass, lumping all types of abnormality together, 3 specific types of abnormality were looked at in detail. These were

  • mucosal pressure contact zones (MCPZ)
  • polyps
  • pus.

Each of these was crosstabulated in the same way, between pain sites and endoscopic sites.

The overall relationship between presence or absence of the abnormality and presence or absence of pain, without regard to site, was also tested by crosstabulation and chi-squared tests.

In cases where a significant association was demonstrated, Goodman and Kruskall’s lambda was calculated, with the site of pain as the dependent variable, as a measure of the value of endoscopic findings in predicting the presence or absence of pain.

Figure 5.1 Pain distribution chart

Facial pain and headache distribution chart used in study

Figure 5.2 Rhinoscopy chart (opens pdf file)

Results

Full data (completed pain charts and endoscopic charts) was available on 161 out of 167 patients.

Details of the endoscopic findings at each site are given in tables 5.1 to 5.6.

Crosstabulated data showed significant associations between pain and endoscopic abnormalities at specific sites (table 5.7).

Significant associations were found between frontal headache and endoscopic abnormalities in the upper part of the nose, including the lateral aspect of the middle turbinate and the ostiomeatal complex (p <0.001).

The pain was located either centrally or on the same side as the abnormality.

Orbital pain and pain between the eyes was similarly associated with endoscopic abnormalities in the upper lateral part of the nose, and with septal spurs.

Mucosal pressure contact zones and the presence of pus showed site-specific associations with pain (tables 5.9 and 5.11) while polyps did not (table 5.10).

Mucosal pressure contact zones lateral to the middle turbinate were associated with frontal and orbital pain.

The presence of pus in the middle meatus was associated with pain over the maxillary sinus on the same side.

The hierarchical log-linear analysis confirmed these significant 2-way interactions between pain and endoscopic sites (table 5.12), with minor differences from the simple crosstabulation analysis (compare with table 5.7).

There were also significant interactions between the presence of abnormalities at one endoscopic site and another. There were significant 3-way interactions involving 2 endoscopic sites and frontal, orbital and malar pain. For pain in the lower jaw, only two-way interactions were significant.

Backward elimination from a saturated model (all effects and all possible interactions between all variables) showed no significant impairment of the model by removing fourth and higher order interactions.

Discussion

These results show that endoscopic abnormalities in the nose are associated with facial pain and headache.

The endoscopic abnormalities are documented in Tables 5.1-6.

Table 5.7 demonstrates statistically significant associations between frontal headache and endoscopic abnormalities in the upper part of the nose. The pain is located either centrally or on the same side as the abnormality. Similar findings apply to orbital pain. Pain between the eyes is associated with endoscopic abnormalities at several sites, both above and below the middle turbinate. This indicates pain may be referred to the central orbital region from widespread afferents. Endoscopic abnormalities below the middle turbinate are also associated with pain in the central malar region and lower jaw.

The statistical significance of these relationships could be challenged on the basis that multiple comparisons are being made. With 72 Chi-squared tests, one would expect 4 or so positive results at the p = 0.05 level by chance alone, and possibly one or two at the p = 0.01 level. However, even using Yates correction, which downgrades significance levels and has been the subject of controversy among statisticians for being too conservative, there are 12 positive associations, including one at the p = 0.001 level (Table 5.7). In all 12 cases, the association is positive, i.e. there is a higher number of patients who have both pain and an abnormality than would be expected by chance alone. If we were looking at random effects, there should be an equal number of significant negative associations, whereas in fact there are none. There is a plausible pathophysiological basis for the positive results, and the overall pattern does not suggest that we are looking at random effects.

Although the statistical significance of the associations does therefore appear to be valid, their predictive value is poor.

Many patients have pain without a corresponding abnormality, while others have abnormalities without a corresponding pain.

To illustrate this point, details of the simple crosstabulation analysis for the strongest association, i.e. pain between the eyes and endoscopic abnormalities lateral to the right middle turbinate, including the ostiomeatal complex, are given in table 5.8.

The Chi-squared residuals for this table show that only 8 out of 99 patients with pain are “explained” by the association. 43 have an abnormality at the site without pain between the eyes, and 9 have an abnormality but no pain.

Although the relationship is highly significant statistically, it is a weak predictor.

Compared with guessing, knowledge of the endoscopic situation will reduce error by 16% (Goodman & Kruskall’s lambda = 0.1613 with pain as dependent variable). This is the strongest association found in the data. The mean value for lambda in the 12 statistically significant associations found is 0.063. This implies that knowledge of the presence or absence of endoscopic abnormalities at these sites will, on average, improve on guessing the site of pain by around 6 or 7%.

Looking at specific abnormalities – mucosal pressure contact zones, polyps and pus – does not improve the crosstabulation results. Polyps were not associated with pain at any site (Table 5.10). Mucosal pressure contact zones had fewer site-specific associations with pain than endoscopic abnormalities in general (Table 5.9), while pus was associated with pain at only one site combination – pain over the antrum when pus was observed in the middle meatus (Table 5.11).

Ignoring site and crosstabulating the presence or absence of pain with mucosal pressure contact zones, polyps and pus reveals no significant associations (Tables 5.12-14). This negative result should, however, be interpreted cautiously. I have already mentioned a caveat on external validity of this study in the introduction to this thesis (see comments on methods and external validity of results). Since 146 out of the 161 patients (90%) had some pain somewhere, crosstabulation measures are not very sensitive, because they are based only on the remaining 15 patients. Furthermore, a significant number of the patients referred were sent because of undiagnosed facial pain and headaches. It is likely that some at least of these patients have no discernible organic basis for their pain, again tending to dilute the strength of any true associations present in the data.

It could be argued that it is not very useful to measure the average strength of this association between subjective and objective findings, since we need to make a decision in each individual case. The main questions in clinical practice are:

a) Does this abnormality cause this patient’s pain?

b) Will surgical correction of the abnormality cure this patient?

If there is pain but no detectable abnormality, the surgical decision is simple because there is nothing to correct. Before dismissing the patient, however, it must be recognised that there may be pathology which has eluded current diagnostic methods. Surgery may even be successful in such cases, acting by an unknown or placebo mechanism. Where the causal relationship is obvious, for instance acute empyema of the sinus, with acute inflammation, purulent secretion and fever, the surgical decision is again simple. The problems arise where the rhinoscopic findings are subtle, and in patients in whom there is reason to believe that the symptoms may be exaggerated, or there may be some other cause for the pain. In order to help with the decision in these cases, it is useful to know the strength of the association between given abnormalities, and pain at specific sites.

This is only part of the picture, and of course it is also helpful to have information on the temporal characteristics of the pain, on associated features, and on patient characteristics suggesting alternative diagnoses such as tumour, migraine, neuralgia, or psychiatric disturbance.

The results of further investigations such as coronal computerised tomography (CT) and magnetic resonance imaging (MRI) can be used (Zinreich et al 1987; Van der Veken et al, 1990; Weber et al 1992; East and Annis 1992). However there is a high incidence of incidental abnormalities on these scans. Reports estimate 30 to 50% of cases (Kennedy et al, 1988; Lloyd et al 1991; Cooke and Hadley 1991; Patel et al 1992). Although it is important to have a scan to exclude any serious pathology, and the scan provides a useful “road map” for endoscopic surgery, the positive predictive value of a scan in attributing headache and facial pain to nasal and sinus causes is seriously limited by this high incidence of incidental abnormalities.

The lack of specificity of CT and MRI has been generally accepted by most ENT surgeons. My study has shown that endoscopic abnormalities may be equally non-specific. Although positive associations have been demonstrated between facial pain and endoscopic abnormalities in the nose, the relationship is weak. That weakness may be due in part to the study population of tertiary referrals.

From a scientific viewpoint, it would be best to study a large population of normal individuals, without pain and headaches, and compare the incidence of nasendoscopic abnormalities with those suffering from pain and headaches of presumed nasal origin. The normal control group would provide an important baseline to assess the significance of endoscopic abnormalities.

Clinically, however, the design of the present study is more apposite. ENT surgeons are not usually asked to give an opinion on asymptomatic normal individuals, we see patients with symptoms. In such a population of patients with symptoms, the above results were found.

Conclusions

Headaches and facial pain are associated with endoscopic abnormalities in the nose.

Frontal headache and pain between and around the eyes are associated with abnormalities in the upper lateral part of the nose, including the lateral aspect of the middle turbinate and the ostiomeatal complex, and with septal spurs.

The pain is located either centrally or on the same side as the abnormality.

This study failed to demonstrate any association between high septal deviations, with middle turbinate contact, and facial pain or headache.

Polyps were not associated with pain at any site.

Although statistically significant relationships were demonstrated, their positive predictive value is low. Many patients have pain without a corresponding abnormality, and many endoscopic abnormalities are found without a corresponding pain.

When a decision has to be made to treat a patient for facial pain and/or headache by correcting intranasal abnormalities, the weakness and variability of these relationships should be taken into account.

Statistical footnote 1: Hierarchical log-linear modelling

Hierarchical log linear modelling is a technique to help determine potentially complex relationships between multiple categorical variables (Norusis, 1988b). It is similar to multiple regression analysis and multivariate analysis of variance, but designed for categorical data. Unlike multiple regression and multivariate analysis, the observations do not have to come from a normal distribution with constant variance.

In this study, I wanted to know the effect of endoscopic abnormalities at six sites on pain at twelve sites. In the initial simple crosstabulation analysis, I carried out 6 X 12 = 72 two-way crosstabulations between single endoscopic sites and single pain sites. This is quite likely to be an oversimplification. For example, if pain in the central orbital region could result as a “final common pathway” from abnormalities either right, left or central in the nose, simple crosstabulations might fail to detect this. There may be additive or other interaction effects between abnormalities at different endoscopic sites. Although 3 dimensional crosstabulation tables can be constructed, they rapidly become difficult to interpret, and beyond 3 dimensions they become almost impossible.

Hierarchical log linear modelling overcomes this problem by constructing an k-dimensional table, where k is the number of variables studied, and taking the natural logarithm of the frequency of cases in each cell of the table. The observed frequencies can then be represented by a linear equation, comprising a term µ – the mean log frequency in all table cells, and a series of parameters lambda, representing the increments or decrements from µ for each individual variable and each possible combination of variables. If this is carried out for all variables and all combinations, the data is completely represented by the linear equation. This is known as a fully saturated model. It fits perfectly, but does not simplify at all.

The next stage is to remove higher-order interactions from the model until removing any more results in a poor representation of the data. The best model is the simplest one which still fits the data reasonably well. It should involve as few higher-order interaction terms as possible. It is arrived at stepwise, by a process of backward elimination of terms from the saturated model.

To decide which terms to eliminate, the goodness-of-fit of the model to the data at each step is examined by a Chi-squared test. If there is a significant difference between the goodness of fit of a model including a higher order interaction and one without, this provides evidence that the interaction is having an effect and that term is retained in the model. The term which least affects the goodness of fit is eliminated, until eventually a stage is reached at which all the remaining terms contribute significantly to the goodness of fit. For a term to remain in the model, the reduction in Chi squared caused by its elimination should not exceed a predefined limit; I used a p value of 0.05 as the elimination criterion.

If the final model, selected according to the predefined criterion, turns out to be a model with main effects only, i.e. no interaction terms, this is evidence that the variables are independent of one another.

Statistical footnote 2: Goodman & Kruskall’s lambda

Lambda is a measure of how well the knowledge of the independent variable predicts the value of the dependent variable. It is designed for ordinal variables and based on proportional reduction in error. It ranges from 0 to 1. Zero means that predicting the dependent variable from knowledge of the independent variable is no better than guessing, i.e. the error is not reduced at all, while 1 means that the error is reduced by 100% i.e. prediction becomes perfect. Lambda of 0.5 means that error is reduced by 50%. Lambda allows for the fact that the dependent variable may already be easy to predict by guessing, for example if nearly all cases fall into one category.