Prediction of peak pressure from clinical and radiological measurements in patients with diabetes

Diabetic patients were selected from the outpatient clinic of the University Hospital Maastricht. Inclusion criteria were diabetes mellitus type 1 (longer than 10 years after date of diagnosis) or type 2 (at least one year after date of diagnosis); age between 30 and 75 years and able to perform daily-life activities without supporting devices. Exclusion criteria were a history of rheumatoid arthritis, severe foot trauma, severe deformity i.e. which require orthopaedic shoes and/or surgery of the foot.

Before the start of the study, patients were informed about all study procedures and their possible risks. The Research Ethical Committee of the University Hospital Maastricht approved the study. One hundred and twenty five eligible patients were screened for peripheral neuropathy through determination of the vibration perception threshold (VPT) at the apex of the hallux with a biothesiometer (Biomedical, Newbury OH)[29, 30]. A VPT higher than 25 Volts was used as the diagnostic criterion for peripheral neuropathy. Table 1 contains descriptive data with respect to gender, age, body mass index (BMI), duration of diabetes and glycosylated haemoglobin (HBa1c). Thirty two patients were excluded because of corrupted plantar pressure data due to technical failure. Finally, data of 93 patients were used for analysis. None of the subjects had a history of foot ulceration.

An EMED SF-4^® pressure sensitive platform (Novel, Munich) was used to quantify the barefoot plantar pressures of the patients' feet and was performed according to a one-step protocol[31, 32]. Barefoot peak pressure was estimated per foot by calculating the mean over the readings of 5 steps. This was done for the whole forefoot and six separate forefoot regions: big toe and metatarsals one (MT-1) to five (MT-5) through weight-bearing anterior-posterior radiographs and Novel 'clinics^®' software[33].

Prior to data collection the grade of callus formation under the plantar aspect of the hallux and the metatarsals' heads was scored according to the method described by Colagiuri et al[17] (1995). The definitions of scores were: grade 1) Distinct area with minimal thickening of keratin layer, grade 2) moderate thickening of keratin layer, grade 3) marked thickening of keratin layer. The callus formation was not removed prior to the data collection, because this was a factor of interest in the prediction of peak pressure.

The 'Valk' scoring system[34] for clinical neurological examination was performed to assess the grade of polyneuropathy. Pinprick sense and light touch sense (cotton wool) of the dorsum of the foot was tested on the mid-foot and compared to the proximal quality of sensation of the ankle. The quality of the vibration sense was tested with a 128 Hz tuning fork. The vibration sense of the big toe was compared to the vibration sense of the ankle, and the vibration sense of the ankle was compared to the patella. The ankle reflex action was compared to the knee reflex action. Pinprick sense, light touch sense, vibration sense of toes and ankles, and ankle reflexes were separately scored for both feet. Criteria for scoring for: normal (0), impaired in comparison with proximal (1) and absent (2), summing up to a maximum score of 20 points. Additionally, light touch sense was related to the anatomical level below which it was impaired: no abnormalities (0), toe (1), mid-foot (2), ankle (3), mid-calf (4) and knee (5). The outcome of the total score could vary between 0 and 25. According to Valk et al (1997) a score higher than 4 was graded as peripheral polyneuropathy[35].

The passive ankle joint dorsal flexion range of motion was measured through use of a plastic goniometer with the patient in prone position and the knee flexed[36]. The range of dorsiflexion of the first metatarsophalangeal joint relative to the first metatarsal shaft was measured in a non-weight bearing position. The bisection lines of the goniometer were placed along the medial shaft of the first metatarsal and the proximal phalanx of the hallux. The degree of hallux valgus was measured in accordance with the guidelines of the American Academy of Orthopaedic Surgeons[37].

The presence of claw toe, mallet toe, curly toe or hammer toe deformity was scored according to the definitions described by Myerson and Shereff[27]. These definitions were used as references for classification of the degree of toe deformity. Accordingly, toe alignment could be normal or deformed as defined by Myerson and Shereff: i.e. 'normal toe' was recorded if there was no deformity present. In addition, less and more malalignment than the definitions described by Myerson and Shereff was arbitrarily classified as minor and severe.

From a dorsal view, a crude judgement of the resting calcaneal stance position was made through visual inspection while the patient was in standing at rest position. The resting calcaneal stance position was recorded as neutral, varus or valgus alignment. In addition, the medial arch height was subjectively classified into: no arch (pes planus); lowered arch; normal arch; elevated arch and excessively elevated arch.

For vascular tests, the patient lay supine and rested for at least 15 minutes prior to testing. A standard examination protocol for each lower extremity was performed. Resting ankle-brachial index (ABI) was obtained through continuous wave Doppler technique and performed at a 45 degree angle to the skin. Ankle-brachial index was calculated with the highest Doppler derived systolic pressure at the ankle (dorsalis pedis or posterior tibialis artery), indexed to the highest Doppler derived brachial artery pressure.

A Hokanson CE-4 strain gauge and photoplethysmograph was used to obtain toe pressure. A suitably sized cuff was wrapped around the proximal phalanx of the hallux. Once the best signal was acquired, the cuff was inflated until the signal disappeared, after which the cuff was slowly deflated until the signal reappeared (Application note: 85315, Hokanson D. E., Inc. Bellevue; WA, USA.). This was taken as the toe systolic pressure. Settings of equipment for all vascular assessment procedures were standardized. All measurements were carried out by the same experimenter. Each measurement was done three times, with the mean used for analysis.

Measurements from standardized static weight-bearing foot radiographs have been shown to be an objective and reliable way of assessing both bony structure and soft tissue dimensions [38‐42]. Weight-bearing anterior-posterior and lateral radiographs were taken with the subject standing on a platform with the central beam for the lateral view (55 kV, 8 mAs) directed horizontally at the plantar aspect of the base of the first metatarsal and the first cuneiform from a distance of 100 cm. For the anterior-posterior view (55 kV, 12 mAs), the beam was directed at the navicular from a distance of 150 cm. This radiographic protocol was previously described by Cavanagh et al[43]. A radio-opaque marker was placed on the radiographic plate in order to judge afterwards whether scale correction was necessary.

All radiographs were taken by the same radiographer, using the same equipment and settings. Measurements were taken through line drawing and a protractor was used to measure the angles. Nine angular and six linear measurements were determined from the lateral radiographs. Five angular and thirteen linear measurements were determined from the anterior-posterior radiographs. All measurements were taken by the same experimenter.

To meet assumptions of normality of statistical distributions peak plantar pressure scores were natural log transformed. At first, data from 186 feet were used for analysis. Repeated measures ANOVA showed no statistical significant differences between left and right feet as a general effect in the repeated measures model, nor specified for the peak pressures scores (i.e. the interactions of peak pressure with left or right side), nor specified for regional peak pressure (i.e. the interactions of regional peak pressure with left or right side). Therefore it was legitimate in the eventual data analysis to average scores over left and right sides.

Next, stepwise regression analysis modelling was performed to determine which of the clinical or radiological measures will explain how much of the variance in averaged local barefoot plantar peak pressure, in each of the six forefoot regions and the whole forefoot separately. At first, all potential predictors of each subsets (clinical or radiological) were entered into the regression model through 'forward selection' and the resulting model with predictors having only statistical significant effects were noted. In the second step, all potential predictors are entered simultaneously and 'backward elimination' was applied until a model was found with predictors having statistical significant effects only. Finally, a cross-section of both resulting models was made using predictors having statistically significant effects in both models and this cross-section of predictors is 'force-entered' into the final regression model[44, 45]. Comprehensive models are assembled from the significant effects of both clinical or radiological predictors following the same three-step procedure. Previously found results from literature were compared with results from our procedure. All regression analysis models were performed using list wise deletion of missing cases. To prevent type I error as much as possible in the multiple use of regression analysis, a Bonferroni correction was applied through the division by the number of plantar regions. Therefore, an alpha level of 0.01 was chosen to judge statistical significance.

To evaluate the prediction models for each individual patient the difference between the predicted value and the actual value was plotted against their mean. This method is also known as residual analysis. In the clinical literature this method is recognized as Bland-Altman analysis, which allows one to judge the agreement between two different measurement techniques[46]. Categorical data were analysed with Chi-square statistics. All data were analysed with SPSS 13.0^® (SPSS Inc., Chicago, USA).

Consistent with what is reported in the literature, higher peak pressures were measured under the neuropathic forefoot than those under the forefoot without neuropathy[51]. The greatest difference was measured in the MT-1 region. The dorsal flexion motion of the metarsalphalangeal joint of the first toe was less while the hallux valgus angle was greater in neuropathic feet compared to feet without neuropathy. Both observations were previously reported in the literature and are positively correlated with plantar pressure[38, 43, 52, 53].

The distributions of callus formation and toe deformity grades were similar for feet with and without neuropathy. Neuropathic feet had a slightly higher grade of deformity of the 4^th and 5^th toe. Results from the regression analysis showed that toe deformities were weakly associated with peak pressures in their pertaining forefoot regions. There were no statistically significant differences in calcaneal resting position and arch height between feet with and without neuropathy. Callosities, toe deformity, calcaneal alignment and arch height were previously identified as distinct features in high risk diabetic patients with neuropathy[1, 49, 54, 55]. The severity of these clinical symptoms in the patients of this study are relatively mild. This could be the explanation for the minor differences between feet with and without neuropathy.

Relatively many differences were found for measurements on lateral radiographs compared to anterior-posterior radiographs. The soft tissue thickness under the head of MT-5 is greater in patients without neuropathy, which could be an explanation for the lower peak pressure in this region compared to patients with neuropathy. A similar relationship between soft tissue thickness and peak pressure exists for MT-1 and was reported in previous studies[5, 38, 56].

Various radiological parameters indicate that neuropathic feet have a lower arch structure than feet without neuropathy. Some parameters obtained from lateral and anterior-posterior radiographs indicate similar classifications found through clinical examination, such as arch height and hallux alignment, although no statistically significant differences were found.

In general, the regression models with selected clinical variables explained more variance in plantar pressure data than the models with only radiological measures, suggesting that clinical examination tells us more about plantar pressure than information from radiographs.

For most regions, the prediction of plantar pressure improved with the combination of radiological and clinical data. The results show that clinical and radiological measurements, either independently or in combination, could explain only about 34 percent of the variance in local bare foot peak pressure. In other words, with the models constructed we have only 34 percent of useful information from the clinical and radiological measurements to make an prediction about the actual bare foot peak pressure.

The Bland-Altman method was used for a visual impression of the individual difference between the predicted and the actual pressure values. We evaluated the comprehensive regression model for the MT-1 region. The scattered data along the wide interval of the horizontal axis in figure 3 show a great variation. The considerable discrepancy between the predicted and the actual peak pressure indicating the inadequacy of the model, especially with regard to the underestimation of high plantar pressure values, i.e. > 700 kPa, for which an accurate estimation is highly important.

In general, the previously reported regression models with independent variables from clinical examination showed similar results to our study. Ahroni et al. (1999) claimed that high in-shoe plantar pressure in diabetic subjects (1,017 feet) could be predicted through the use of clinical variables, such as body mass, insulin use, Caucasian race, male sex, plantar callus and diabetes duration[49]. Only 12 percent of the variation was explained. Although Ahroni in his study found body mass to be identified as the most important factor, Cavanagh et al. (1991) found that body mass was a poor predictor[57].

Various predictive models relating plantar peak pressure and structural variables obtained through imaging techniques have been suggested)[38, 47‐49, 58‐62]. Cavanagh et al. presented a model which explained approximately 40 percent of the variance in forefoot pressure during walking with 3 to 4 static structural measurements from foot radiographs of healthy subjects[38]. While we found a similar result, 34 percent of explained variance is not sufficient for prediction of plantar pressure.