Single Sensor Gait Analysis to Detect Diabetic Peripheral Neuropathy: A Proof of Principle Study

Article information

Diabetes Metab J. 2018;42(1):82-86
Publication date (electronic) : 2018 January 03
doi :
1Movement Science Group, Oxford Brookes University, Oxford, UK.
2Oxford Institute of Nursing & Allied Health Research, Oxford, UK.
3Oxford Centre for Diabetes, Endocrinology & Metabolism, Churchill Hospital, Oxford University Hospitals NHS Foundation Trust, Oxford, UK.
4NIHR Oxford Biomedical Research Centre, Oxford, UK.
5Department of Clinical Neurology, University of Oxford, Oxford, UK.
Corresponding author: Johnny Collett. Movement Science Group, Oxford Brookes University, Oxford, OX3 0BP, UK.
Received 2017 April 29; Accepted 2017 October 16.


This study explored the potential utility of gait analysis using a single sensor unit (inertial measurement unit [IMU]) as a simple tool to detect peripheral neuropathy in people with diabetes. Seventeen people (14 men) aged 63±9 years (mean±SD) with diabetic peripheral neuropathy performed a 10-m walk test instrumented with an IMU on the lower back. Compared to a reference healthy control data set (matched by gender, age, and body mass index) both spatiotemporal and gait control variables were different between groups, with walking speed, step time, and SDa (gait control parameter) demonstrating good discriminatory power (receiver operating characteristic area under the curve >0.8). These results provide a proof of principle of this relatively simple approach which, when applied in clinical practice, can detect a signal from those with known diabetes peripheral neuropathy. The technology has the potential to be used both routinely in the clinic and for tele-health applications. Further research should focus on investigating its efficacy as an early indicator of or effectiveness of the management of peripheral neuropathy. This could support the development of interventions to prevent complications such as foot ulceration or Charcot's foot.


Diabetic peripheral neuropathy (DPN) affects over 50% of people with diabetes mellitus [1]. DPN can lead to reduced independence, lower rates of perceived health and higher mortality rates [12] as well as a range of complications including neuropathic ulceration, Charcot's feet, and eventual lower limb amputation. DPN itself often begins years before a formal diagnosis is made [2]. Timely identification of DPN enables the introduction of preventative management, education, and advice measures, which can reduce the rate of the complications of DPN. However, many individuals are not diagnosed with DPN promptly, and hence fail to receive timely intervention [3].

The effect DPN has on gait is highlighted in recent systematic reviews [45]. However, whilst the reviews demonstrate temporal-spatial gait parameters are significantly affected by DPN [4], gait analysis does not have sufficiently utility to be used routinely in primary clinical settings. This may be partly due to most objective gait analysis reliant on technology such as pressure mats, specialised treadmills, or laboratories with optical motion capture and force measurement, being time, space and resource intensive and thus a barrier to routine practice [6]. Inertial measurement units (IMUs) are relatively cheap, robust, and easy to use, and also contained within smart phones, opening the possibility for a telehealth application [7]. The data obtained from IMU can be used to estimate temporal-spatial gait parameters and provide sensitive data indicative of gait deficits found in various neurological pathologies [78910]. However, these methodologies have yet to be explored in a primarily DPN pathology.

We propose that gait screening may have the potential to aid in the detection and monitoring of DPN, supporting interventions to prevent complications associated with the diabetic foot, which could increase mobility function and quality of life as indicated by results in other neurological conditions [1112]. Due to the novelty of the proposed gait assessment mythologies in this area, we set out to test the principle that single, torso-mounted, IMU gait analysis can differentiate those with DPN from age and body mass index (BMI)-reference data in the clinical setting.



People were recruited from the Oxford Centre for Diabetes, Endocrinology & Metabolism (OCDEM) at the Oxford University Hospitals NHS Foundation Trust (Oxford, UK). Inclusion criteria were (1) a clinical diagnosis of DPN (diagnosed on the basis of clinical history and physical examination including tests of light touch and vibration sensation in line with international guidelines [13]); (2) failure of the monofilament test (10 and 75 g); and (3) a score of ≥8 on the Rivermead mobility index and able to walk independently with or without walking aid. Exclusion criteria were (1) a history of amputation or active ulcers (those with previously healed ulcers were eligible); (2) insufficient capacity to consent or uncontrolled psychiatric symptoms; (3) confounding gait factors such as neurological conditions, knee replacements, amputations, or padding; and (4) pregnancy. The study was conducted according to the Declaration of Helsinki and received National Research Ethics Committee approval (NRES: 11/SC/0218).

Reference control data were extracted from a database held by Oxford Brookes University (BIG Database, Oxford, UK). The database consists of data from 2,030 individuals aged between 5 and 80 years with no known pathology affecting gait primarily obtained from visitors to the Science Museum London from April to July 2012 (UREC: 100490). The data base can be interrogated to provide reference data according to desired demographic. Healthy ageing gait performance has been well described in the literature generally resulting in slower gait [14] and reduced gait symmetry [15], with similar results also reported for those with higher subcutaneous thigh fat [16] as well as for DPN [4]. Therefore, based on available demographic data age (REFAGE) and BMI (REFBMI) reference data was extracted from the database, whereby all individuals within 1 standard deviation of age and BMI of the DPN group were included.


After obtaining informed consent, demographic data of gender, age, height, weight, shoe size, leg length, time since diabetes diagnosis, Rivermead mobility index, and Barthel index were obtained.

Gait analysis for both groups was performed during a standard 10-m test by attaching an IMU to the lower back that sampled triaxial accelerometer and gyroscope data at a frequency of 100 Hz. Assessment for those with DPN took place at OCDEM in a corridor free from obstacles. Further methodology on how gait parameters were derived has been previously extensively described [89101718].

Spatiotemporal parameters derived were step time (ms), cadence (steps/min), stride length (m), and walking speed (m/s).

Gait control parameters calculated were beta (deg), SDa (au), SDb (au), ratio (au) [10], and walk ratio (mm/[steps/min]) which have been described as indicators of neuro motor control [19].

Statistical analysis

Group differences were assessed by means of independent samples t-test or chi-square statistic. Discriminatory power was further investigated using receiver operating characteristic (ROC) curves, by means of the area under the curve (AUC). Statistical analysis was performed using, compare means, cross tabs, and ROC curve procedures in SPSS version 22 (IBM Co., Armonk, NY, USA).


Descriptive data

Seventeen people (14 men) with DPN and 42 (30 men) controls were included in analysis 63.2±9.2 years (mean±SD). There was no difference in age, gender ratio, height, or BMI between groups (P>0.05) (Table 1). DPN was confirmed by none of the participants responding to the 75 or 10 g monofilament test performed by a specialist diabetes podiatrist [20]. Baseline glycosylated hemoglobin was 8.8%±1.0% (73.3±11.5 mmol/mol) and all patients had evidence of microvascular disease (pre-proliferative retinopathy or urinary microalbuminuria). The majority (eight people) scored 14 on the RMI which indicated little impairment to functional mobility, but not able to run or walk fast without a limp.

Descriptive data

Gait analysis

Comparison between DPN and control groups can be found in Table 2, differences were found for all spatiotemporal and control variables except for stride length (P<0.05). Walking speed produced the greatest discriminatory power (AUC=0.975), with good discriminatory power (AUC >0.8) also found for step time and SDa and fair discriminatory power (AUC >0.7) found for cadence, beta, and ratio.

Gait assessment data


This proof of principle study found, for the first time that a single IMU used in the clinical setting during a 10-m walk has the potential to discriminate those with DPN compared to healthy walking.

We intentionally selected those with established DPN in order to determine if we could detect the spatiotemporal gait deficits previously described in this population and found our results were similar to those reported in recent reviews [45]. Interestingly, in our DPN cohort, with minimal functional mobility deficits, walking speed was the most sensitive parameter and average walking speeds in the DPN group was just above the 0.8 m/sec cut-off indicative of frailty in the elderly [21]. Furthermore, this reduction in walking speed can be primarily attributed to temporal rather than spatial changes, as no significant difference were found in stride length compared to control. A previous study, also using a single lower back mounted IMU to measure gait, in those with diabetes but excluding those with DPN, found the reduction in walking speed compared to control was primarily due to spatial gait changes, as they found no significant difference in cadence but a significantly reduced stride length [22]. Whilst, we cannot generalise from this proof of principle study these results are consistent with the results of a recent meta-analysis that gait deficits differ between people with diabetes with and without DPN [5].

Non-linear gait measures may offer further insight into gait dynamics in diabetes than revealed by traditional spatial temporal measurement. Khalaf et al. [23] derived gait entropy from the plantar pressure wave forms during stance and found an entropy was difference between those with DPN and those with diabetes without complication. In the current study we employed a non-linear approached that utilizes the dynamic similarity of the periodic wave forms produced by the centre of mass during walking, that can easily and rapidly be obtained through placing an IMU on the lower back of the participant. Whilst, this is the first time this methodology has been applied in DPN, it has have been previously shown utility in neurological populations in determination of gait control parameters that have shown to be sensitive to subtle gait disturbances in movement disorders [89]. The ‘ratio’ parameter has been found to differentiate those with the Huntington's gene but not displaying symptoms from healthy control walking [9] and in Parkinson's disease SDa was found to account for differences with healthy walking [10]. In the current study, we found SDa had good discriminatory power, with beta and ratio parameters having ‘fair’ discriminatory power. These the results are encouraging that DPN may present with a characteristic signature and warrant further investigation in those with less severe or pre-symptomatic DPN. In addition these gait measures in combination with assessments of the severity of the neuropathy such as monfiliment, vibration pressure threshold, and biothersiometer techniques may improve our understanding of the functional effects of the pathology and support more efficient diagnosis and monitoring.

Our results demonstrate that a relatively cheap, easy to use single sensor approach can differentiate those with severe DPN over a short-distance walking test when compared to age and BMI matched reference data. Indeed whilst, the potential benefits of large scale gait analysis ‘in the field’ and reference databases have been proposed, to date no suitable methodology has been established [24]. A simple methodology, such as the one presented, might assist with identification of those at risk and facilitate earlier diagnosis DPN resulting in a more efficient management. Future studies are therefore indicated and should focus more on people at risk off or with mild DPN. Longitudinal studies may provide insights into interindividual changes and define parameters which identify when a person with diabetes has clinically relevant neuropathy and the utility of the technology for clinicians ‘in the field’ and for telehealth applications.


Dax Steins received funding by the higher education innovation fund (HEIF) to conduct the above research. Patrick Esser and Helen Dawes were funded by the Elizabeth Casson Trust. In addition, Patrick Esser received funding by the EPSRC, Helen Dawes by the National Institute for Health Research (NIHR) Oxford Biomedical Research Centre and Johnny Collett by the Health Education England Thames Valley (HEETV). Garry D. Tan was supported by the National Institute for Health Research (NIHR) Oxford Biomedical Research Centre based at Oxford University Hospitals NHS Trust and University of Oxford. This paper presents independent research supported by the NIHR. The views expressed are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health.

We would like to thank the patients and staff of the Oxford Centre for Diabetes, Endocrinology & Metabolism.


CONFLICTS OF INTEREST: Oxford Brookes University hold UK, EU, and USA Patent (PCT/GB2009/051767) on the methods used to transpose and integrate data inertial data. Patrick Esser, Johnny Collett, and Helen Dawes are inventors on patent.


1. Hoffman EM, Staff NP, Robb JM, St Sauver JL, Dyck PJ, Klein CJ. Impairments and comorbidities of polyneuropathy revealed by population-based analyses. Neurology 2015;84:1644–1651. 25832668.
2. Callaghan B, Kerber K, Langa KM, Banerjee M, Rodgers A, McCammon R, Burke J, Feldman E. Longitudinal patient-oriented outcomes in neuropathy: importance of early detection and falls. Neurology 2015;85:71–79. 26019191.
3. Tesfaye S. Recent advances in the management of diabetic distal symmetrical polyneuropathy. J Diabetes Investig 2011;2:33–42.
4. Mustapa A, Justine M, Mohd Mustafah N, Jamil N, Manaf H. Postural control and gait performance in the diabetic peripheral neuropathy: a systematic review. Biomed Res Int 2016;2016:9305025. 27525281.
5. Hazari A, Maiya AG, Shivashankara KN, Agouris I, Monteiro A, Jadhav R, Kumar S, Shashi Kumar CG, Mayya SS. Kinetics and kinematics of diabetic foot in type 2 diabetes mellitus with and without peripheral neuropathy: a systematic review and meta-analysis. Springerplus 2016;5:1819. 27812455.
6. Toro B, Nester CJ, Farren PC. The status of gait assessment among physiotherapists in the United Kingdom. Arch Phys Med Rehabil 2003;84:1878–1884. 14669198.
7. Steins D, Sheret I, Dawes H, Esser P, Collett J. A smart device inertial-sensing method for gait analysis. J Biomech 2014;47:3780–3785. 25305689.
8. Esser P, Dawes H, Collett J, Howells K. IMU: inertial sensing of vertical CoM movement. J Biomech 2009;42:1578–1581. 19442978.
9. Collett J, Esser P, Khalil H, Busse M, Quinn L, DeBono K, Rosser A, Nemeth AH, Dawes H. Insights into gait disorders: walking variability using phase plot analysis, Huntington's disease. Gait Posture 2014;40:694–700. 25172806.
10. Esser P, Dawes H, Collett J, Howells K. Insights into gait disorders: walking variability using phase plot analysis, Parkinson's disease. Gait Posture 2013;38:648–652. 23510514.
11. Hauser RA. Early pharmacologic treatment in Parkinson;s disease. Am J Manag Care 2010;16(Suppl Implications):S100–S107. 20297870.
12. Kremens D, Hauser RA, Dorsey ER. An update on Parkinson's disease: improving patient outcomes. Am J Med 2014;127:S3.
13. Schaper NC, Van Netten JJ, Apelqvist J, Lipsky BA, Bakker K;. IWGDF). Prevention and management of foot problems in diabetes: a summary guidance for daily practice 2015, based on the IWGDF guidance documents. Diabetes Res Clin Pract 2017;124:84–92. 28119194.
14. Rosano C, Studenski SA, Aizenstein HJ, Boudreau RM, Longstreth WT Jr, Newman AB. Slower gait, slower information processing and smaller prefrontal area in older adults. Age Ageing 2012;41:58–64. 21965414.
15. Aboutorabi A, Arazpour M, Bahramizadeh M, Hutchins SW, Fadayevatan R. The effect of aging on gait parameters in able-bodied older subjects: a literature review. Aging Clin Exp Res 2016;28:393–405. 26210370.
16. Beavers KM, Beavers DP, Houston DK, Harris TB, Hue TF, Koster A, Newman AB, Simonsick EM, Studenski SA, Nicklas BJ, Kritchevsky SB. Associations between body composition and gait-speed decline: results from the Health, Aging, and Body Composition study. Am J Clin Nutr 2013;97:552–560. 23364001.
17. Esser P, Dawes H, Collett J, Feltham MG, Howells K. Assessment of spatio-temporal gait parameters using inertial measurement units in neurological populations. Gait Posture 2011;34:558–560. 21764583.
18. Esser P, Dawes H, Collett J, Feltham MG, Howells K. Validity and inter-rater reliability of inertial gait measurements in Parkinson's disease: a pilot study. J Neurosci Methods 2012;205:177–181. 22269595.
19. Rota V, Perucca L, Simone A, Tesio L. Walk ratio (step length/cadence) as a summary index of neuromotor control of gait: application to multiple sclerosis. Int J Rehabil Res 2011;34:265–269. 21629125.
20. Nazhel B, Yetkin I, Irkec C, Kocer B. Sympathetic skin response in diabetic neuropathy. Electromyogr Clin Neurophysiol 2002;42:181–185. 11977432.
21. Castell MV, Sanchez M, Julian R, Queipo R, Martin S, Otero A. Frailty prevalence and slow walking speed in persons age 65 and older: implications for primary care. BMC Fam Pract 2013;14:86. 23782891.
22. Awotidebe TO, Ativie RN, Oke KI, Akindele MO, Adedoyin RA, Olaogun MO, Olubayode TE, Kolawole BA. Relationships among exercise capacity, dynamic balance and gait characteristics of Nigerian patients with type-2 diabetes: an indication for fall prevention. J Exerc Rehabil 2016;12:581–588. 28119881.
23. Khalaf K, Al-Angari HM, Khandoker AH, Lee S, Almahmeed W, Al Safar HS, Jelinek HF. Gait alterations in the UAE population with and without diabetic complications using both traditional and entropy measures. Gait Posture 2017;58:72–77. 28756345.
24. Brown S, Boulton A, Bowling F, Reeves N. Benefits, challenges, and potential utility of a gait database for diabetes patients. J Diabetes Sci Technol 2016;10:1065–1072. 27022098.

Article information Continued

Funded by : higher education innovation fund
Funded by : Elizabeth Casson Trust
Funded by : Engineering and Physical Sciences Research Council
Funded by : National Institute for Health Research

Table 1

Descriptive data

Demographic DPN (n=17) Control (n=42) Deltaa (P value)
Age, yr 63±9 (46–81) 61±4 (54–72) 2±2 (0.307)
Sex, male:female 14:3 30:12 X2 (0.883)
Height, m 1.78±0.9 (1.56–1.91) 1.77±0.9 (1.49–1.93) 0±0 (0.882)
BMI, kg/m2 33.6±7.6 (24.1–51.7) 31.6±3.9 (25.1–44.6) 2.0±1.9 (0.320)
Time since diagnosis diabetes 24±13 (3–49) - -
Barthel index 19±1 (15–20) - -
RMI 13±2 (9–15) - -

Values are presented as mean±standard deviation (range).

DPN, diabetic peripheral neuropathy; BMI, body mass index; RMI, rivermead mobility index.

aDelta (between group difference; DPN-Control) reported with P value of independent samples t-test or for nominal data P value for chi-sqaure statistic reported.

Table 2

Gait assessment data

Gait parameter DPN Control P valuea ROC-AUC (95%CI) ROC P valueb
 Step time, msec 650±117 543±47 0.002 0.804 (0.657–0.950) >0.001
 Stride length, m 1.33±0.16 1.41±0.17 0.066 0.632 (0.466–0.799) 0.114
 Cadence, steps/min 95±16 112±10 0.001 0.797 (0.650–0.944) >0.001
 Walking speed, m/sec 0.85±0.23 1.34±0.18 >0.001 0.975 (0.943–1.000) >0.001
 Beta 27±10 34±10 0.014 0.735 (0.600–0.870) 0.005
 SDa 1.98±0.36 3.91±1.99 >0.001 0.849 (0.748–0.950) >0.001
 SDb 0.56±0.14 0.91±0.56 >0.001 0.667 (0.533–0.800) 0.046
 Ratio 3.67±0.94 4.86±2.30 0.044 0.782 (0.649–0.914) 0.001
 Walk ratio 7.2±1.1 6.4±1.0 0.012 0.696 (0.551–0.841) 0.019

Values are presented as mean±standard deviation.

DPN, diabetic peripheral neuropathy; ROC-AUC, receiver operating characteristic area under the curve; CI, confidence interval.

aIndependent samples t-test probability value, coefficient reported, with 95% confidence intervals, bReceiver operating characteristic probability value.