Predicting diabetic foot ulceration using routinely collected data in a foot clinic. What level of prognostic accuracy can be achieved?

This study aimed to investigate the efficacy of using routinely collected clinical data in predicting the risk of diabetic foot ulcer (DFU). The first objective was to develop a prognostic model based on the most important risk factors objectively selected from a set of 39 clinical measures. The second objective was to compare the prediction accuracy of the developed model against that of a model based on only the 3 risk factors that were suggested in the systematic review and meta‐analyses study (PODUS). In a cohort study, a set of 12 continuous and 27 categorical data from patients (n = 203 M/F:99/104) who attended a specialised diabetic foot clinic were collected at baseline. These patients were then followed‐up for 24 months during which 24 (M/F:17/7) patients had DFU. Multivariate logistic regression was used to develop a prognostic model using the identified risk factors that achieved p < 0.2 based on univariate logistic regression. The final prognostic model included 4 risk factors (Adjusted‐OR [95% CI]; p) in total. Impaired sensation (116.082 [12.06–1117.287]; p = 0.000) and presence of callus (6.257 [1.312–29.836]; p = 0.021) were significant (p < 0.05), while having dry skin (5.497 [0.866–34.89]; p = 0.071) and Onychomycosis (6.386 [0.856–47.670]; p = 0.071) that stayed in the model were not significant. The accuracy of the model with these 4 risk factors was 92.3%, where sensitivity and specificity were 78.9%, and 94.0% respectively. The 78.9% sensitivity of our prognostic 4‐risk factor model was superior to the 50% sensitivity that was achieved when the three risk factors proposed by PODUS were used. Also our proposed model based on the above 4 risk factors showed to predict the DFU with higher overall prognostic accuracy. These findings have implications for developing prognostic models and clinical prediction rules in specific patient populations to more accurately predict DFU.


| INTRODUCTION
Diabetic Foot Ulcer (DFU) is the main cause of non-traumatic lower limb amputation worldwide.
The lifetime prevalence of DFU is reported to be between 15% and 25% 1 in a person with diabetes, while a recent study estimates this figure to be higher. 2 Nearly half of the diabetic foot ulcers become infected 3 and one in five moderate or severe diabetic foot infections lead to amputation. 4,5 It has been reported that the presence of DFU increases the risk of death at 5 years by 2.5 times. 6 Systematic reviews of existing literature identify many predictive factors including impaired sensation, peripheral vascular disease, peak plantar pressure, foot deformities and fasting blood sugar as risk factors for diabetic foot ulcers. 7,8 Also, age, duration of diabetes, height, body weight and Body Mass Index (BMI) have been associated with the risk of DFU occurrence. 7,8 An earlier systematic review of the risk factors for diabetic foot ulceration identified foot deformity, peripheral neuropathy (Vibration Perception Threshold-VPT or cutaneous insensitivity to monofilament), peripheral arterial disease (pulses and/or ankle brachial index), previous amputation, the presence of callus, HbA1c, Tinea pedis, and onychomycosis as prognostic factors that can predict the risk of ulceration. 9 In studies focusing on validation and comparison of existing diabetic foot risk models, it was concluded that the existing models show high efficacy represented as the area below receiver operating characteristic (ROC) ranging from 0.73 to 0.86 and with no significant difference in accuracy between them. 10 However, in a later multicentre prospective cohort study the authors reported considerable differences in the efficacy of predictions when they applied to a hospital versus community settings i.e. the area below the ROC curve could differ between 0.46 in a community setting for a risk prediction model and 0.86 for another model. 10,11 Recently, the prognostic factors for foot ulceration in people with diabetes were investigated as part of the international research collaboration for the prediction of diabetic foot ulcerations known as PODUS. 12 This investigation proposed a multivariable prognostic model based on a systematic review and meta-analysis using individual patient data from 10 studies to predict foot ulceration. 12 In PODUS, the history of DFU, insensitivity to a 10-g monofilament and any absent pedal pulse were identified as consistent independent predictors of DFU. 12 This 3 risk-factor prognostic model proposed in PODUS was reported to have a sensitivity of 90.0%-95.3% and a specificity of 12.1%-63.9%. 12 This prognostic model was reported to compare favourably with the more complex approaches to foot risk assessment recommended in clinical diabetes guidelines. 12 However, the data in those studies were collected during a period no later than 2008 within a mix of settings from the hospital to primary care, outpatients, and tertiary care units. 12 This indicates that although this 3-risk factor prognostic model proposed by PODUS compares favourably with more complex models in general, 12 the variability which was previously reported for existing risk models across different settings 11 can also exist. These indicate that there is a need for assessing the efficacy of using routinely collected data in predicting the risk of DFU in the first place and to compare the accuracy of such against that of proposed by the 3-risk factor PODUS prognostic model.
Hence, this study aimed to assess the efficacy of using routinely collected clinical data at a foot clinic setting in predicting DFU. The first objective was to develop a prognostic model based on the most important risk factors that are objectively selected from a set of routinely collected clinical data. The second objective was to compare the prediction accuracy of the developed model against that of a model based on only the 3 risk factors that were suggested in PODUS. 12

| MATERIALS AND METHODS
Data from a routine NHS clinic in England was analysed as a part of this work. This audit received necessary governance approvals prior to any data reduction and synthesis. Data from patients (n = 203, M/ F:99/104) who attended the specialised diabetic foot clinic were included in this study. The primary exclusion criterion was the presence of any DFU at baseline. Diabetic foot ulcer was defined as a full-thickness wound involving the foot or the ankle, distal to and including the malleoli. The sample size was calculated using logistic regression with a power of 0.95 and Alpha 0.05, Odds Ratio of 1.8 for future ulceration. This resulted in a sample size of 203 participants who would also provide 80% power to detect hazard ratios of 1.6.
An initial database that included 211 patients was identified.
Eight patients had missing data and were removed from the data set and the participants with complete and no missing data set were 203 patients who were all included in the study.  Tables 1 and 2. The primary inclusion criteria were that the patient was diagnosed with diabetes.
The categorical and continuous parameters assessed are shown in Tables 1 and 2. In addition, data related to previous ulceration, and amputation along with foot-specific characteristics such as muscle weakness and foot deformity were also assessed. Twenty-five out of 203 participants in the study had a history of previous ulceration (Table 2).
Skin status was considered as Dry when the epidermis lacked moisture or sebum; and Normal when the skin was well-balanced eudermic that is neither too oily nor too dry. 13 Tinea pedis and onychomycosis 14 and the presence of callus were recorded. 15,16 A 10-g monofilament was used on 10 sites for each foot and sensation in less than 8 out of 10 sites was considered as neuropathy. 17 The vibratory and blunt sensations were tested over the tip of the great toe bilaterally and the abnormal response was defined as when the patient loses sensation. 1,18 The foot-specific categorical parameters for each participant were defined as if these occurred on either or both feet for each participant. The presence of triphasic, biphasic, monophasic, or  based on the univariate analyses in which variables with p < 0.2 were selected. 19 There were a total of 20 parameters where the Univariate analyses resulted in p < 0.2 added to the multivariate mode. These included 4 continuous parameters (as highlighted in Table 1) and 16 categorical parameters (as highlighted in Table 2)

| RESULTS
All results related to the categorical and continuous measures are presented in Tables 1 and 2

| Differences in patients with DFU and with no-DFU during follow-up
The present study highlights that the patients with future DFU occurrence had distinctive characteristics in a set of parameters, majority of which were related to an impaired sensation that is in line with the previous studies 11 in European, Middle Eastern 20 and African population. 21 In addition, a significantly higher proportion of patients who incurred future DFU had a history of ulceration (Large effect size), which is in line with the findings in the European 10,11 Middle Eastern 20  The results of the present study also highlight that the group vulnerable to future diabetic foot ulcers have significantly higher HbA1C and creatinine levels, which is in line with the previous finding in the European 11 population.
In addition, the results of the current study also indicate that a significantly higher proportion of participants who developed a future ulcer were males, which is in line with the previous findings in the Middle Eastern population. 20 However, these results of our study is contrary to the previous studies on the European population showed no association between sex and future DFU. 10,11 We also found that the group with future DFU was significantly heavier (with a large effect size), which contradicts the results of a previous study in which no significant differences in weight were found for those with future DFU in the European 10,11 Middle Eastern 20 and African 21 populations.

| Independent risk factors associated with increased likelihood (odds) of future diabetic foot ulcer
The associations between weight and HbA1C with increasing the likelihood of DFU occurrence found in the present study are in line with a study conducted on the pooled patient data from Europe and North America. 12 Also, in the current study, male gender and retinopathy were associated with increased likelihood (OR:2.873 and OR:2.778 respectively) of future DFU that is in line with the results of pooled patient data from Europe and North America (OR:1.69 and OR:2.09 respectively). 12 In this study muscle wasting was also found to significantly increase the likelihood of future ulceration (OR:13.125) in line with our The 78.9% sensitivity achieved here in our study is much higher than the 50% based on 3 risk factors suggested by the PODUS model. 12 However, this is still low, where at least 1 in 5 patients with future DFU are missed in the prediction. In the future, the inclusion of mechanical properties of plantar soft tissue 24 should be considered, which could increase the model sensitivity.
When the average risk of DFU at 2 years of follow-up was calculated based on the three risk factor PODUS model, 12 the NAEMI ET AL.
predicted risk was 11.87 � 20.36%. The risk predicted by the pro-