The 10‐year follow‐up of a community‐based cohort of people with diabetes: The incidence of foot ulceration and death

Abstract Background Identifying people with diabetes who are likely to experience a foot ulcer is an important part of preventative care. Many cohort studies report predictive models for foot ulcerations and for people with diabetes, but reports of long‐term outcomes are scarce. Aim We aimed to develop a predictive model for foot ulceration in diabetes using a range of potential risk factors with a follow‐up of 10 years after recruitment. A new foot ulceration was the outcome of interest and death was the secondary outcome of interest. Design A 10‐year follow‐up cohort study. Methods 1193 people with a diagnosis of diabetes who took part in a study in 2006–2007 were invited to participate in a 10‐year follow‐up. We developed a prognostic model for the incidence of incident foot ulcerations using a survival analysis, Cox proportional hazards model. We also utilised survival analysis Kaplan–Meier curves, and relevant tests, to assess the association between the predictor variables for foot ulceration and death. Results At 10‐year follow‐up, 41% of the original study population had died and more than 18% had developed a foot ulcer. The predictive factors for foot ulceration were an inability to feel a 10 g monofilament or vibration from a tuning fork, previous foot ulceration and duration of diabetes. Conclusions The prognostic model shows an increased risk of ulceration for those with previous history of foot ulcerations, insensitivity to a 10 g monofilament, a tuning fork and duration of diabetes. The incidence of foot ulceration at 10‐year follow‐up was 18%; however, the risk of death for this community‐based population was far greater than the risk of foot ulceration.


| INTRODUC TI ON
Data from the International Diabetes Federation (IDF) shows 537 million adults worldwide have a diagnosis of diabetes mellitus in 2022, and this is predicted to rise to 643 million by 2030. 1 The complications of diabetes can cause premature death and considerable morbidity for people who have a diagnosis of diabetes.
Complications affecting the lower limb include vascular and sensory impairment (ischemia and neuropathy) both of which can result in foot ulceration, soft tissue infections and ultimately lower limb amputations.There are high healthcare costs associated with these outcomes and managing complications which affect the lower limb requires the greatest proportion of healthcare expenditure for people with diabetes. 2entifying those who are likely to experience a foot ulcer is an important part of preventative care with national and International Diabetes Clinical Guidelines traditionally recommending that annual foot risk assessments are conducted to categorise a person's risk of developing a foot ulcer as either low/moderate or high. 3,46][7] Even more rare are reports of patient outcomes followed up over the longer term.

More recently, analyses of patients' routinely collected data have
shown the risk of death is considerably higher than that of developing a foot ulcer for people with diabetes and people with diabetes who experience foot ulcers have also been found to have a greater risk of death than those who do not. 8,9 author of this manuscript previously published the results of a cohort study conducted between 2006 and 2008 which aimed to quantify the predictive value of elements of clinical history, diagnostic test results and symptoms and signs for foot ulceration in the general diabetes population recruited in a community healthcare setting. 10People registered with the NHS Tayside podiatry service in Scotland gave consent for their health data to be collected and analysed (n = 1193) and were followed up for an average period of 11 months between 2007 and 2008.The average age of the participants was 70 years, there were almost equal numbers of men and women, and the average duration of diabetes was almost 9 years.These demographic features were comparable to the wider diabetes population in Scotland.Only 23 participants experienced a foot ulcer within the original follow-up period giving rise to concerns about the accuracy of the model. 10As part of a wider research project, we sought additional consent from the participants of the original cohort with diabetes to conduct a long-term follow-up of outcomes. 11

| The study aims and objectives
We aimed to develop a predictive model for foot ulceration in diabetes using the incidence of first foot ulcerations 10 years after recruitment.A new foot ulceration was our outcome of interest and death was the secondary outcome.
The primary objectives were to observe the incidence of foot ulceration (outcome) in the cohort of people with diabetes over a 10-year period and to develop a predictive model for foot ulceration in this group based on 25 explanatory variables collected in the original study considered to be the most readily available from patient records held by interdisciplinary healthcare professionals. 10,12,13The secondary objective was to observe the rate of mortality in the cohort population over a 10-year period.A description of the tests is provided in the Box 1.

| Participants and consent
The original cohort included 1193 consecutively recruited participants from NHS Tayside community podiatry clinics. 10Those with a diagnosis of diabetes mellitus who were ambulant and free of foot ulceration at the time of recruitment gave informed consent and had a detailed examination by one of eight podiatrists to collected baseline explanatory variables (2006 and 2007) (Box 1).The first follow-up to ascertain the presence of the primary outcome (foot ulceration) was performed on average 11 months after recruitment by podiatrists who scrutinised the hand-held records of all people who took part for the occurrence of a foot ulcer.It was assumed that the population might receive standard foot care during the conduct of the study but no specific data about this were collected.
For the 10-year follow-up study, the subject of this manuscript, participants of the original study were identified via their Community Health Index (CHI) number on an electronic database (SCI Diabetes).
Having identified those who had died in the intervening period, a postal letter invited those who survived to participate in the follow-up study and give consent.Outcome data (foot ulcers) were collected by a podiatrist working in NHS Tayside and ascertained from NHS Tayside patient podiatry records.The NHS podiatrist was unaware of the original risk status of participants as determined in the original cohort study.For the development of the prognostic model, we included information from the entire original cohort.For those that had died in the intervening period and also those that did not respond to a request for consent, we only included their data

Ankle brachial index (ABI)
Patients were in a resting state and their feet level with their hips for at least 20 min before this test was performed.A sphygmomanometer blood pressure gauge (Speider & Keller) was used to measure blood pressure at the arm and ankle.A doppler ultrasound transducer was used to detect a posterior tibial, anterior tibial or brachial pulse.Where the ankle pressures exceeded 220 mmHg blood pressure measurement was abandoned.Ankle pressure was divided by arm pressure to give a ratio, <0.8 was regarded as indicative of ischemia and >1.3 potentially indicative of arterial calcification.

HbA1c
Routinely collected data were obtained from an electronic source (Scottish Clinical Information Diabetes Care).Three HbA1c measurements were used to produce an average HbA1c reading for each patient.A reading of 7.5% and less was regarded at target HbA1c and more than 7.5% was regarded as poor blood glucose control.

Monofilament
A 10g filament Semmes Weinstein (SWF) was placed at 90° to the foot and pressure applied until the filament bent.Patients were asked if they could feel the touch of the filament on the 1st, 2nd, 4th, 5th met head and apex of the 3rd toe.Inability to feel the touch with a monofilament in either foot was regarded as a positive test result.

Neurothesiometer
The voltage was turned up full (50 volts) to allow the patient to feel the vibration on the palm of their hand.The dial was then turned down to zero, the probe was placed against the medial MPJ and the voltage turned up slowly until the patient could feel vibration.
This was repeated three times on each side.Vibration readings of >25 volts were considered as a positive test result.

Neurotip™ (www. owenm umford. com)
Patients were allowed to feel both the sharp and blunt end of the neurotip on their index finger or the dorsum of the foot before being asked to close their eyes.The blunt end of the neurotip on the plantar aspect of the hallux was always used and patients asked whether they perceive it to be sharp or blunt.Inability to distinguish between sharp and blunt was considered a positive test result.

Temperature
The podiatrist placed the two flat end surfaces of a TIP-Therm® rod on the dorsum of each foot.With their eyes closed patients were asked whether it felt cold or not so cold.Inability to distinguish between cold and not so cold was regarded as a positive test result.

Tendon hammer
In a standing position with one knee on a static chair with the Tendo Achilles (TA) clearly visible, the tendon hammer was used to tap the patient's TA.A brisk plantar flexion was judged as a pass and an absent plantar flexion regarded as a positive test result.

Tuning fork
An un-calibrated tuning fork was vibrated on the podiatrist own thigh (to reduce the noise) and placed over the medial aspects the 1st MPJ.Patients were asked whether they could feel the vibration.Inability to feel the vibration was regarded as a positive test result.

Cotton wool
Patients were asked to say if they could feel light touch when cotton wool was lightly rubbed over the dorsum of the foot.Inability to feel light touch was regarded as a positive test result.

Peak plantar pressure
Peak plantar pressure data were collected using the Pressure Stat system™ manufactured by Podotrack (www.footl ogic.info) Patients placed one foot on a single pressure mat.Readings were interpreted by a clinician blind to the results of all other tests and 0-1.5 kg/ cm 2 was the threshold above which (>1.5 to 15 kg/cm 2 ) peak plantar pressure was regarded as abnormally high.
of follow-up data due to death or lack of consent was modelled using a right censoring approach. 14

| Sample size
As the size of the sample analysed in this follow-up study depended on the originally collected dataset and the number of consents, we could only perform a retrospective sample size validity calculation based on the final number of events and model predictors, which can be found in the results section. 15

| Cox proportional hazards model
Cox proportional hazards modelling was used to develop a predictive model for foot ulcerations over a 10-year follow-up period (average 129 months).For the modelling purposes, only the first occurrence of foot ulcerations recorded either during the initial or the 10-year follow-up period were taken into consideration.We followed participants from their individual date of entry (in 2006-2007) up to a first ulceration, censoring for death and either the end of the initial study period for those who did not provide consent or November 2017 for those who provided additional consent for the 10-year follow-up study.Time to event was calculated in months.
The survival analysis was censored for death (n = 489) and either the end of the initial study period for those who did not provide consent (n = 277) or November 2017 for those who provided additional consent for the 10-year follow-up study (n = 311).
Thus, censoring accounted for the fact that everyone included in the initial study provided consent up to the end of the initial study period.Only those subjects with additional consent for the follow-up study are followed up to the end of the follow-up study period.
Univariate Cox proportional hazards models were utilised for all candidate variables in the data set for pre-selection purposes.
Clinical experts were included in discussions about the availability of variables from patients' health records and these discussions also influenced the selection of candidate variables for the multivariable analysis. 7No Bonferroni-type adjustment was made for the significance level of the univariate pre-selection tests.Although we aimed to reduce the number of variables considered for the initial multivariate model due to sample size and power considerations, we wanted to do so conservatively, pre-selecting variables even if they were associated with moderate evidence for significance.The multivariate model was developed using a backwards selection algorithm beginning with a model that contained all the significant variables To measure the performance of the selected model, the receiver operating characteristic-area under the curve (ROC-AUC) was calculated using the Chambless and Diao's (2006) estimator of AUC for time-to-event data. 16

| Competing risk analysis
Competing risk analysis was performed using the cumulative incidence function (CIF) with death as a competing event for foot ulceration. 17,18

| Kaplan-Meier analysis of survival
Further analysis using Kaplan-Meier curves was performed for each binary explanatory variable included in the Cox proportional hazards model, to compare the difference in survival time (all-cause mortality) between the two groups.The log-rank test was performed to validate the significance of that difference.

| RE SULTS
Patient data from the SCI Diabetes database showed that at the 10-year follow-up, 489 participants (41%) had died, and 116 participants were lost to follow-up, leaving 588 participants classified as alive and able to be contacted for their consent (Figure 1).Of the 588 participants who survived, 311 (53%) gave consent for their health records to be accessed.We obtained ethical approval and Caldicott approval from NHS Tayside to ascertain the outcome of 50 deceased participants whose podiatry records were not yet destroyed.Outcome data at 10-year follow-up was available for a total of 361 participants, of whom a total of 65 experienced a foot ulceration at 10 years.
Demographic details of people who survived who did not give consent to follow-up were older, more likely to be female, had diabetes for a shorter duration and fewer ulcers than those who did give consent.(Tables 5 and 6).
The values for the exposure variables collected during the original screening process at the time of recruitment to the study were used as the exposure variables.Table 1 provides the main demographic characteristics of the cohort.Of the 23 participants who developed a foot ulcer after ~1 year in 2008, 16 of those died during the 10-year follow-up period.

| Retrospective sample size calculation
Based on our final Cox model, after backwards selection, we calculated a minimum sample size requirement n = 597 for a model with 4 parameters corresponding to the 4 predictor variables.
The required number of Events Per Parameter is 11, as calculated by following the approach in Riley et al. 15 The complete case data set available for developing the multivariate Cox proportional hazards model (with 4 parameters) satisfies the minimum sample size requirements, as it includes observations with information censored due to death or lack of consent after the end of the initial study, that is it includes 1032 observations and 77 foot ulcers (events). 15where the value for the last 3 covariates is 1 when there is insensitivity to a 10-g monofilament, inability to feel a tuning fork, and previous history of ulceration.Otherwise, it is 0. Also, h i (t), is the hazard function, that is the estimated probability that subject i experiences an ulceration at time t + 1, conditional on the fact that they have not experienced one at time t.Additional information on the model fit relevant to prediction is given in the Appendix S1. 19 The summarised ROC-AUC at 10 years for the final Cox proportional hazards model was 0.732 (95% CI: 0.674-0.8). 18This shows an increased risk of ulceration for those with previous history of foot ulcerations, as well as insensitivity to monofilament and tuning fork.The risk for ulceration also increases with diabetic duration.

| Competing risk analysis
The cumulative incidence of death was calculated as a competing event for foot ulceration.At 10 years, the cumulative incidence for F I G U R E 2 Competing risk analysis: cumulative incidence of death and foot ulcers.
Table 4 provides descriptive statistics for the exploratory variables selected in the multivariate model by all-cause mortality status of the participants at 10-year follow-up.For those who exhibit Insensitivity to either 10 g monofilament or VPT tuning fork, we observed a higher percentage with a 'Dead' status compared and those with a history of foot ulceration.For those subjects, the survival probability decreases faster over time than for people who are not insensitive to a 10 g monofilament or tuning forks or have experienced a foot ulceration (Figures 3-5).

| CON CLUS IONS
The risk factors for foot ulceration in this population identified by our multivariable survival analysis are consistent with validated prognostic models and clinical prediction rules for foot ulceration in international datasets, and in this study are shown to be sustained over the long term. 7,11An inability to feel a 10 g monofilament or the vibration of a tuning fork underlines the central importance of neuropathy in the development of foot ulceration and their inclusion in the recommendations of diabetes clinical guidelines are justified. 3,4Previous foot ulceration is well-established as an independent predictor of foot ulceration risk, but it does signify advanced disease and is therefore of limited use in prevention.to distinguish between cold and cool temperatures were not found to be independently predictive of foot ulceration in this new analysis. 10The increase in statistical power from the larger number of foot ulcers (from n = 23 to 77) has produced a predictive model with good discrimination; the C statistic (AUC) being 0.73 (95% CI: 0.674-0.805).This C statistic occupies the lower end of the confidence interval of the original predictive model (0.835 (95% CI 0.735 to 0.936)) and although there is no statistically significant difference between the two models, there may be residual confounding building up over the long follow-up period.
Fifty-three per cent of the original cohort who survived a further 10 years, gave consent to have their data collected and analysed.The observed long-term incidence of foot ulceration of 18% in this cohort is consistent with foot ulcer incidence reported by others who have estimated the lifetime risk of foot ulceration in people with diabetes to be 25%. 20However, there is uncertainty about the true rate of ulceration in this cohort in the long term due to missing data for F I G U R E 5 Survival probability over Time for subjects with previous ulcerations (ulcerhistory = Yes) and subjects without previous ulcerations (ulcerhistory = No).The p-value is obtained after testing for a significant difference between the two groups with the log-rank test.
those who were deceased, lost to follow-up or who did not provide consent for their long-term outcomes to be collected.In any observational study of risk, there is a possibility that people will receive preventative interventions and the ulceration rate of this cohort may have been modified after general podiatric care.
By comparing demographic characteristics of those who died and those who survived and those who gave consent and those who did not, we have explored possible explanations for missingness.
(Tables 5-8) The profile of those who died compared with those who did not, shows those who died were older, had diabetes for longer and a greater number had a previous history of foot ulceration than those who survived and indicates the natural history of diabetes in this cohort population.(Tables 7 and 8).
A comparison of the demographic profiles of those who gave consent to have their long-term follow-up data analysed and those who did not shows the consenters were slightly younger, had diabetes for longer, and more had a previous history of foot ulceration, concern about which may have acted as an incentive to participate in the research.(Tables 5 and 6) The reluctance of those who did not agree to the 10-year follow-up may stem from the fact fewer had experienced a foot ulcer and the study objectives may have been perceived to be less relevant to those individuals.
up to the end of the first 2006-2008 study, for which consent was received.By utilising a Cox proportional hazards model, the absence BOX 1 Description of diagnostic tests (p-value <0.05) from the univariate analysis.The test of significance for each variable retained in the final multivariate Cox model produced a p-value ≤.05.F I G U R E 1 Flow diagram describing the progress of the cohort at the 10-year follow-up.Outcome data available n = 50 26/50 people developed foot ulcers before they died Consented for data collecƟon at 10Y Follow up N = 311 39/311 people developed foot ulcers Did not consent for 10Y Follow-up data to be Collected n = 277

F I G U R E 3
to the 'Alive' percentage.Both p-values from the corresponding chi-square tests of association for the two 2 × 2 cross-tabulations are <10 −6 , a significant result that shows evidence of an association between Alive/Dead status and Sensitivity/Insensitivity to monofilament or VPT tuning fork.There is also a statistically significant association between Ulceration History and Alive/Dead status (p-value = .04).Kaplan-Meier survival curves, with death as the outcome, show a statistically significant smaller survival probability for subjects that demonstrate insensitivity to a 10 g monofilament or a tuning fork Survival probability over Time for subjects with Insensitivity to 10 g Monofilament (monofilament = Yes) and subjects without Insensitivity (monofilament = No).The p-value is obtained after testing for a significant difference between the two groups with the log-rank test.

F I G U R E 4
Survival probability over Time for subjects with Insensitivity to VPT tuning fork (VPTtuningfork = Yes) and subjects without Insensitivity (VPTtuningfork = No).The p-value is obtained after testing for a significant difference between the two groups with the logrank test.This long-term follow-up predictive model differs from the original predictive model in that ankle brachial index, insulin use in the 3 months prior to recruitment, previous amputation and an ability

Table 2
Characteristics of the cohort at baseline.Exploratory variables included in the univariate analysis with foot ulceration as the outcome variable.
7rovides the results of univariate Cox proportional hazards models for 26 potential risk factors of foot ulceration.Of the 26 variables tested in the univariate Cox models, we found 14 variables that reached statistical significance at p < .05.These variables were also identified during previous discussions with an international group of authors of cohort studies examining the risk of foot TA B L E 17The complete case data set available for multivariate analysis of the 14 variables included 1032 individuals with an average duration of diabetes of 8.79 years (SD: 8.12), 507 (49.13%) females and 77-foot ulcerations.The final multivariate Cox proportional hazards model included: duration of diabetes (years) [Hazard Ratio (HR): 1.039, 95% 95% CI: 1.409-3.712]andprevious history of foot ulceration [HR: 2.564, 95% CI: 1.404-4.682](Table3).The fitted model is, log h i (t) h 0 (t) = 0.038 × (Dur. ) + 1.008 × (In.Mon. ) + 0.827 × (In.Tun.Fork) + 0.941 × (Hist.), TA B L E 2 TA B L E 3 Multivariate Cox proportional hazards model results.
Note: 'Dead' relates to all-cause mortality: cross-tabulation between Insensitivity to Monofilament and Alive/Dead status, between Insensitivity to VPT tuning fork and Alive/Dead status and between Previous History of Ulceration and Alive/Dead status.Also, mean (standard deviation) of Time from diagnosis of diabetes for Alive vs. Dead participants.
Demographic data for participants who were alive and consented to follow-up.Demographic data for participants who were alive but did not consent to follow-up.Demographic data of participants who died during the follow-up and 10-year foot ulceration were data available.Demographic data of participants who died during the follow-up for whom 10-year foot ulceration data were not available.Demographic characteristics of participants who were lost to follow-up for whom 10-year foot ulceration data are not available.
TA B L E 5 TA B L E 6 TA B L E 7 TA B L E 8 TA B L E 9