figure b

Introduction

Diabetes affects one in 11 people globally and is the leading cause of non-traumatic lower extremity amputation (LEA) [1,2,3]. Population-based studies have reported large variations in the prevalence and incidence of LEAs in people with diabetes, in part related to the different recording periods and inconsistencies in the methods used to define LEA and diabetes [4,5,6]. In a previous hospital-based study conducted in Hong Kong, 53.3% of people who underwent LEA had coexisting diabetes, and in the USA, appropriately 75% of all LEAs were attributed to diabetes [7, 8].

Apart from causing major physical disability, LEA has a profound impact on the psychological well-being and social function of the affected individual [2, 3, 9]. Furthermore, LEA is a strong predictor of short-term mortality. A recent meta-analysis of studies conducted between 2005 and 2015 reported that up to half of the people died within the first year after LEA [10]. Given the rising prevalence of diabetes and the improvement in life expectancy of people living with diabetes over the past two decades, the number of people at risk of developing long-term complications including LEA is expected to also increase [11]. Reliable epidemiological data on trends related to LEA and associated mortality are useful to assess whether the quality of care for people with diabetes has improved and to inform health policy makers and healthcare providers of where additional resources may be required to fill the care gaps. Relevant studies from Asia are scarce due to lack of a reliable and representative data source [4, 5]. The newly established territory-wide Hong Kong Diabetes Surveillance Database (HKDSD) offers a unique opportunity to monitor and evaluate clinical outcomes of people with diabetes in Hong Kong at a population level. Although we previously reported declining trends in overall LEA rates in people with diabetes in Hong Kong, the changing patterns in relation to anatomical level and the trends in mortality after LEAs are still unclear [12]. In this study, we describe the trends in the rates of hospitalisation for LEAs stratified by the anatomical level of LEA and the 1 year all-cause mortality rates after LEA among people with diabetes in Hong Kong between 2001 and 2016.

Methods

Data source and study population

Hong Kong is a special administrative region of China with a population of 7.4 million. The Hong Kong Hospital Authority (HA) is a statutory body established in 1990 that governs all 43 public hospitals and 122 specialist/general outpatient clinics, providing about 90% of total health services in Hong Kong [11]. In 2000, the HA adopted an electronic medical record (EMR) system to capture healthcare information of all people attending public hospitals and clinics including demographic data, diagnostic and procedure codes, laboratory tests and drug prescriptions. The HKDSD is a territory-wide population-based database of people with diabetes identified from the HA EMR system. Detailed information about the study design and the population of the HKDSD has been reported previously [11, 13]. Briefly, from the HA EMR system, people with diabetes were identified based on: (1) ICD-9 code 250; (2) revised edition of the International Classification of Primary Care, World Organization of National Colleges, Academies, and Academic Associations of General Practitioners/Family Physicians code T89 or T90; (3) HbA1c level of 48 mmol/mol (6.5%) or greater; (4) fasting plasma glucose level of 7.0 mmol/l (126 mg/dl) or greater; or (5) prescription of blood glucose-lowering drugs or long-term insulin (≥28 days). All people in the HKDSD aged 20 years or older were included in this study. This study was approved by the local clinical research ethics committee.

Ascertainment of LEA and 1 year mortality after LEA

We identified hospitalisations for LEAs using ICD-9 procedure codes listed in the hospital discharge records. We defined minor LEAs as any LEA performed below the ankle (ICD-9 procedure codes: 84.11 [toe] and 84.12 [foot]) [7]. We defined major LEAs as any LEA performed at the ankle or above (ICD-9 procedure codes: 84.13–84.16 [ankle to knee] and 84.17–84.19 [above knee]). We excluded discharges with a traumatic amputation diagnosis code (ICD-9 codes: 895–897). We only included the highest-level LEA when two or more LEA procedure codes were recorded during the same hospitalisation to avoid overestimation, because we were not able to assess whether these LEAs were revisions after initial amputation, unilateral or bilateral, or planned multi-step procedures [7, 14]. A very small number (n = 29, <0.3% of total LEA events) of LEAs without the level specification (ICD-9 code: 84.10) were not included in the study. All people in the HKDSD were followed up for mortality from linkage to the Hong Kong Death Registry. Death in the HKDSD was only available for year and month. We defined 1 year mortality after LEA based on the vital status at 12 months after hospital discharge for LEA. Deaths that occurred during the hospital stay were included in the 1 year mortality.

Statistical analysis

We limited the analyses to data from 2001 to 2016 to avoid bias from incomplete case records of diabetes and LEAs in the first year of establishment of the Hong Kong HA EMR system. We performed all analyses separately for minor LEAs and major LEAs: (1) because they have different operation aims and are associated with different economic costs, functional implications and survival prospects [6]; and (2) for ease of comparison with other studies. We calculated annual event rates of LEAs in people with diabetes as the number of hospitalisations for LEAs divided by the mid-year population in the HKDSD, and further stratified by sex and age (20–64, 65–74 and ≥75 years). We used the number of hospitalisations for LEAs as the numerator as it measured both disease severity and healthcare costs [15]. We calculated annual cumulative 1 year mortality rates after LEA by dividing the number of deaths within 1 year after LEA by the number of LEA events. All rates were age standardised to the 2016 Hong Kong Census mid-year population. The 1 year mortality rate analyses were restricted to 2001 to 2015 as at least 1 year after LEA was required to define the 1 year mortality events. Age-stratified analyses were not performed for 1 year mortality rates due to the small number of deaths. We calculated the age-standardised annual hospitalisation rates for gangrene, ulcer, cellulitis/abscess and peripheral artery disease (PAD), which are major risk factors for diabetes-related LEAs [2]. The age-standardised annual rates of hospitalisation for lower-limb revascularisation procedures were also determined. The ICD-9 diagnosis and procedure codes used to identify the hospitalised events are shown in the electronic supplementary material (ESM) Table 1. We have previously reported changes in HbA1c and LDL-cholesterol control among all people in the HKDSD [11]. Here, we additionally reported trends in mean HbA1c and LDL-cholesterol in the year of LEA and trends in the proportion of people who had hospitalisations for cardiovascular diseases (including ischaemic heart disease, acute myocardial infarction, heart failure and stroke) within 1 year prior to LEAs among people who had undergone LEAs.

We used Joinpoint regression models to describe whether there were significant trends in rates of hospitalisation for LEAs and 1 year mortality after LEA, and to assess whether the trends changed during the study period [16]. The Joinpoint regression model uses a Monte Carlo permutation test to detect the time points (joinpoints) at which significant changes in trend occur (either the direction or magnitude) [16]. The analysis compares models by starting with no joinpoints and subsequently testing whether joinpoints are needed to be added into the model to best fit the data. The best-fitting model is selected to report the annual per cent change (APC) in each linear segment detected and the average annual per cent change (AAPC) in the entire study period. The AAPC is equal to the APC if the trend is linear in the entire study period.

Sensitivity analysis

Increasing rates of diabetes detection through improved screening efforts can enlarge the pool of people with shorter diabetes duration over time, causing decline in LEA rates even when the actual risk of developing complications has not changed. To address potential lead time bias, we performed sensitivity analysis excluding people with newly diagnosed diabetes in each study year from the denominator. Additionally, to remove the effect of multiple LEAs on 1 year mortality rates, we performed sensitivity analysis excluding people who had more than one LEA within each study year. Although we could not distinguish the type of prevalent diabetes, we have previously developed and validated an algorithm to delineate incident type 1 from type 2 diabetes between 2002 and 2015 in the HKDSD [13, 17]. We performed sensitivity analysis excluding people with incident type 1 diabetes (0.4% of all incident diabetes) to assess the effect of different types of diabetes on trends in LEA rates and 1 year mortality rates. We performed all analyses using R (version 3.5.3, Vienna, Austria) or the Joinpoint Regression Program (version 4.7.0.0; Statistical Methodology and Applications Branch, Surveillance Research Program, National Cancer Institute, Bethesda, MD, USA). A two-tailed P value less than 0.05 was considered statistically significant.

Results

Between 2001 and 2016, a total of 390,071 men and 380,007 women with diabetes aged 20 years or older were included in the HKDSD, among whom 6113 hospitalisations for LEAs in men and 4149 hospitalisations for LEAs in women were recorded (Table 1). Major LEAs contributed to 54.0% and 66.3% of all LEAs in men and women, respectively. Ninety-four per cent of all minor LEAs were performed at the toe in both sexes, whilst 60.5% of major LEAs in men and 46.6% in women were performed above the knee (ESM Fig. 1). Men were younger than women at the time of undergoing LEAs (Table 1).

Table 1 Numbers of hospitalisations for LEAs and 1 year deaths after LEA in men and women with diabetes in Hong Kong between 2001 and 2016
Full size table

Trends in rates of LEAs by sex

During the whole study period, men had a 2.3-fold (95% CI 2.1, 2.4) higher risk of minor LEAs and 1.6-fold (95% CI 1.5, 1.6) higher risk of major LEAs than women after age adjustment. The hospitalisation rates for LEAs fell dramatically in both sexes with diabetes, with the decline being more marked in major than minor LEAs (Fig. 1, ESM Table 2). From 2001 to 2016, the age-standardised rates of minor LEAs declined by 48.6% (AAPC: −3.8; 95% CI −5.7, −1.9) from 14.0 (95% CI 10.4, 17.7) to 7.2 (95% CI 5.8, 8.6) per 10,000 in men and by 59.5% (AAPC: −6.3; 95% CI −10.6, −1.8) from 7.9 (95% CI 2.3, 13.6) to 3.2 (95% CI 1.5, 4.9) per 10,000 in women. The rates of major LEAs declined by 77.9% (AAPC: −8.0; 95% CI −9.6, −6.5) from 19.5 (95% CI 15.5, 23.4) to 4.3 (95% CI 3.5, 5.1) per 10,000 in men and by 79.3% (AAPC: −10.4; 95% CI −13.1, −7.6) from 11.6 (95% CI 9.4, 13.9) to 2.4 (95% CI 1.7, 3.1) per 10,000 in women. All of the declining trends were linear except for the trend in major LEAs in women, in whom the decline was greater in 2001–2006 (APC: −16.4; 95% CI −22.1, −10.2) than in 2006–2016 (APC: −7.2; 95% CI −10.6, −3.6). The larger decrease in rates of major than minor LEAs resulted in a significant increase in the proportion of minor LEAs among all LEAs from 38.4% to 54.4% in men and from 32.1% to 41.5% in women from 2001 to 2016 (ESM Fig. 1). Sensitivity analysis excluding people with newly diagnosed diabetes (ESM Fig. 2) or incident type 1 diabetes (data are not shown) yielded similar declining trends in rates of LEAs. Despite the decline in the hospitalisation rates for LEAs, the absolute number of LEAs increased except for major LEAs in women (Table 1). The number of revascularisations among people with diabetes was small and increased over time in men and stabilised in women (ESM Table 3). The age-standardised rates of revascularisation declined in both men (AAPC: −2.7; 95% CI −4.7, −0.8) and women (AAPC: −10.5; 95% CI −14.3, −6.6) with diabetes. There were no significant changes in mean HbA1c levels over time among men and women who had undergone minor or major LEAs, but a slight decline was observed for mean LDL-cholesterol levels over time (Table 1). In addition, no clear trend was observed for the proportion of amputees who had hospitalisations for ischaemic heart disease, acute myocardial infarction, heart failure or stroke within 1 year prior to undergoing LEAs (ESM Table 4).

Fig. 1
figure 1

Age-standardised rates of hospitalisation for minor (a, b) and major (c, d) LEAs in men (a, c) and women (b, d) with diabetes in Hong Kong between 2001 and 2016. Dots are observed age-standardised rates of hospitalisation for LEAs; lines are modelled age-standardised rates of hospitalisation for LEAs from Joinpoint regression models

Full size image

Trends in rates of LEAs by age

The linear declining trends in rates of LEAs were also observed across all age subgroups, most notably in older people (Fig. 1). This has led to a narrowing of both absolute and relative differences (rate ratios) in age-standardised rates of LEAs between young and middle-aged (20–64 years) and the oldest (≥75 years) subgroups over time (ESM Table 5). An increase in the contribution of young and middle-aged people (20–64 years) to hospitalisations for LEAs was detected, from 43.0% to 52.7% in men and from 14.3% to 29.5% in women having minor LEAs, and from 26.0% to 39.0% in men and from 10.3% to 17.2% in women having major LEAs, between 2001 and 2016 (Fig. 2).

Fig. 2
figure 2

Proportions of minor (a, b) and major (c, d) LEAs in men (a, c) and women (b, d) with diabetes by age group in Hong Kong between 2001 and 2016

Full size image

Trends in rates of gangrene, ulcer, cellulitis/abscess and PAD

In total, 10,924, 10,443, 28,616 and 8843 hospitalisations for gangrene, ulcer, cellulitis/abscess and PAD, respectively, were recorded in people with diabetes between 2001 and 2016, and the rates were consistently higher in men than women. Significant declines in hospitalisation rates were observed for gangrene, ulcer and PAD, but not for cellulitis/abscess (Fig. 3). The decline was pronounced for PAD, which fell by 67.1% (APC: −22.6; 95% CI −28.6, −16.1) in men and by 74.6% (APC: −18.5; 95% CI −29.2, −6.3) in women from 2001 to 2006, but then flattened until 2016 in both sexes. When stratified by age subgroups, the decline in hospitalisation rates for gangrene, ulcer and PAD was generally greater in the older (65–74 years and ≥75 years) than younger (20–64 years) subgroups in both sexes (ESM Table 6).

Fig. 3
figure 3

Age-standardised rates of hospitalisation for gangrene (a), ulcer (b), cellulitis and abscess (c) and PAD (d) in men and women with diabetes in Hong Kong between 2001 and 2016. Dots are observed age-standardised rates of hospitalisation; lines are modelled age-standardised rates of hospitalisation from Joinpoint regression models

Full size image

Trends in 1 year mortality rates after LEA

Between 2001 and 2015, 18.5% of men and 21.3% of women died within 1 year after minor LEAs, and the corresponding proportions of people dying after major LEAs were 41.8% in men and 42.0% in women (Table 1). Cardiovascular disease, pneumonia, infections and renal disease were the most common causes of death, with little difference between sexes, or between minor and major LEAs (ESM Table 7). After age standardisation, the rates of 1 year mortality after major LEAs (30.3 [95% CI 25.3, 35.3] per 100 in men and 29.0 [95% CI 23.8, 34.2] per 100 in women) were almost threefold higher than the rates after minor LEAs (11.4 [95% CI 8.8, 14.0] per 100 in men and 10.3 [95% CI 6.7, 13.8] per 100 in women) during the whole study period, but no significant difference existed between men and women. The Joinpoint regression models showed that the 1 year mortality rates after both minor and major LEAs remained constant during the surveillance period in both sexes (Fig. 4). The AAPCs for 1 year mortality rates were 1.5 (95% CI −3.0, 6.1) in men and 0.6 (95% CI −4.7, 6.1) in women after minor LEA, and −2.1 (95% CI −4.8, 0.6) in men and 2.6 (95% CI −1.4, 6.7) in women after major LEA. Sensitivity analyses excluding people with multiple LEAs within 1 year (ESM Fig. 3) or incident type 1 diabetes (data are not shown) resulted in similar non-significant changes in 1 year mortality rates.

Fig. 4
figure 4

Age-standardised 1 year mortality rates after minor and major LEAs in men (a) and women (b) with diabetes in Hong Kong between 2001 and 2015. Dots are observed age-standardised 1 year mortality rates after LEA; lines are modelled age-standardised 1 year mortality rates after LEA from Joinpoint regression models

Full size image

Discussion

Our study is one of the few studies describing trends in the rates of diabetes-related LEAs in Asia, and contributes to current gaps in the literature on the global epidemiology of LEAs in people with diabetes [5]. Using population-level surveillance data, we revealed a 50–80% reduction in rates of hospitalisation for LEAs among people with diabetes in Hong Kong over a 16 year period, with the decline greater for major than minor LEAs and in the elderly than the young. Corresponding decreases in the rates of gangrene, ulcer and PAD were detected. However, there was no improvement in 1 year mortality rates after minor or major LEAs in both sexes.

Declining trends in rates of LEAs

It is estimated that up to 85% of diabetes-related LEAs are avoidable [3]. The encouraging downward trends in rates of LEAs may reflect improvements in health behaviours such as smoking cessation [18] and control of cardio-metabolic factors, including blood glucose levels, blood cholesterol levels and blood pressure [11, 19]. Most importantly, patient education, increased awareness and appropriate management of diabetic foot were instrumental in preventing LEA, through prevention of foot ulcers and gangrene. The introduction of a territory-wide complication screening programme initiated in hospital-based diabetes centres in 2000, which later expanded to become the Risk Assessment and Management Programme (RAMP) implemented in primary care clinics since 2007, provides regular assessment of complications and facilitates early detection of feet at risk as well as fast-tracked referral to podiatrists and surgeons for people with diabetes [19, 20]. It has been reported that RAMP participants had a 50% lower risk of foot ulcers or LEAs than the propensity score-matched non-RAMP group during a median follow-up of 3 years [21]. We also observed significant decline in rates of PAD and corresponding decline in rates of revascularisation in both sexes. Although vascular intervention has been considered an important procedure to improve limb salvage rates and reduce LEAs in Western countries, its use is less frequent in Chinese populations [22, 23]. This is in part due to lower prevalence of PAD among Chinese compared with other ethnic groups, accounted for by differences in upstream risk factors and inherent susceptibility [18, 24]. The reduction in LEA rates may be also due to changes in other characteristics of people with diabetes, including earlier detection, resulting in an increasingly healthier background population with diabetes with fewer comorbidities and better metabolic profiles [13, 19].

Comparison across different studies is limited by the lack of uniformity in the methods used to identify LEAs in previous studies. For instance, LEA events could be recorded as the number of LEA-related procedure codes, the number of hospitalisations for LEAs, or the number of amputees or first-time amputees [5, 15]. Furthermore, variations in defining the population with diabetes could also affect the rate estimates. Despite these methodological issues, most studies from Western countries and a few Asian countries have reported declines in the rates of LEAs in people with diabetes from the early 1990s to the late 2000s, with generally greater decline in major than minor LEAs [5]. Recently, an increase of diabetes-related LEAs was detected in the USA, predominantly among the young and middle-aged group aged 18–64 years, raising concerns that preventive efforts were not reaching certain subgroups of people with diabetes who may have different disease characteristics and needs [7]. In the present analysis, we did not identify any joinpoints that could indicate a change in the otherwise linear declining trends. Despite the territory-wide Hong Kong population coverage, the results of our study could not be generalised to other parts of China or other Asian countries, given dissimilarities in healthcare systems and the demographic and socioeconomic characteristics of populations [25]. However, the successful changes in healthcare delivery which might have contributed to the declining trends in rates of LEAs in Hong Kong may serve as a reference for other countries where the current incidence of LEAs is high.

Minor vs major LEAs and young vs older people

Major LEAs contributed to more than half of all LEAs in our study, which was similar to findings for Scotland and Taiwan [26, 27], but higher than for other Western countries such as the USA, England and Belgium [7, 28, 29]. The differences in the proportions of major vs minor LEAs could be accounted for by differences in the methods of determining LEA level. The higher percentages reported in our study could also reflect delay in patient recognition and surgical intervention for foot problems when compared with other countries, so increasing the risk of progression from initial minor LEAs to major LEAs or the need for initial major LEA surgery. The smaller improvements in rates of minor LEAs than major LEAs, and in young and middle-aged people than the elderly, need careful interpretation. Minor LEAs have been considered an important surgical strategy to limit the progression of foot problems in order to avoid major LEAs [6, 30]. The results in our study might reflect more aggressive diabetic foot management over time, with earlier minor LEAs at a younger age to prevent later more serious major LEAs and functional disability at an older age. Conversely, they could be due to poorer diabetes management in young people [31]. Consistent with emerging data on the worse prognosis of young-onset vs usual-onset diabetes, we have previously reported less improvement in control of blood glucose and cholesterol in younger than older people and an unchanged mortality rate in young people aged 20–44 years in Hong Kong [11]. Young people with diabetes have less motivation for diabetes self-care against other life priorities and are also more vulnerable to diabetes-related distress and other mental health problems, which adversely affect disease control and increase their downstream risk of developing complications [31,32,33,34]. The decision to amputate and the timing of LEAs are also influenced by medical culture, surgical techniques and doctors’/patients’ attitudes. Future studies are needed to clarify whether these factors have changed during the study period to better interpret our findings.

Unchanged 1 year mortality rates after LEA

Crude cumulative mortality rates after minor and major LEAs at 1 year were about 20% and 40%, respectively, in our study, and are comparable with previous publications [10]. The absence of improvement in mortality rates after LEA was observed not only in our study but also in other countries [35]. The high and unchanged 1 year mortality rates after LEA in people with diabetes are concerning. It is conceivable that people who had LEAs were often severely ill with multiple morbidities which put them at a high risk of developing other life-threatening complications. This is evident by the high percentages of amputees being hospitalised for major cardiovascular events within 1 year before LEA shown in our study. However, the present database lacks information on events leading up to the LEA or other factors around treatment pathways to fully explain the observed trends.

Strengths and limitations

Major strengths of this study are the long surveillance period and the territory-wide population coverage, although <10% of people who presented to the private sector might be not included in our study. Our study has several limitations. First, our results are based on an administrative database with potential issues of validity, despite a previous systematic review showing high quality and accuracy of hospital discharge codes in administrative data [36]. Second, establishing an accurate prevalence of people with diabetic foot diseases and PAD is difficult, as many events or procedures occurred in outpatient clinics and were not captured in the current database. We only included hospitalised events with foot diseases and PAD as the principal diagnoses and identified presentations at the more severe end of the spectrum. Third, comparison between people with and without diabetes could not be made. As a result, it is unclear to what extent the decline in LEA rates in people with diabetes could be attributable to improvements in diabetes care or overall healthcare delivery. Fourth, we were not able to distinguish between the first and repeated LEAs as the medical history of people in the HKDSD before 2000 was not available. Finally, we were only able to distinguish the type of diabetes for a subset of incident cases in the HKDSD [13, 17]. However, the very small number of people with type 1 diabetes in Hong Kong should have little effect on the results [32, 37].

Conclusions

In conclusion, despite substantial decline in hospitalisation rates of LEAs in people with diabetes in Hong Kong, 1 year mortality rates after LEA were high and remained constant over a 16 year period. Continuous efforts including earlier foot management, better control of cardio-metabolic risk factors and appropriate surgical intervention are needed to further prevent LEAs and improve the survival rate of people undergoing LEAs.