Background
Lower limb amputation is a common consequence of advanced peripheral vascular disease, often secondary to the long-term consequences of diabetes [
1]. Unfortunately, little is known about the number people living with limb loss given the paucity of prevalence data and uncertainty inherent in estimating prevalence based on historical trends in amputation incidence and mortality [
2]. By comparison, there is a comparatively large body of literature that suggests the incidence rate of lower limb amputation has remained fairly constant over the last 15 years [
3,
4]. A more detailed look at these data suggests there may have been a shift in the types of lower limb amputations performed [
3‐
7]. The incidence rate of transtibial amputation (TTA) seems to have declined [
3,
8‐
11], while there is some evidence that the incidence rate of partial foot amputation (PFA) has increased proportionately [
3,
7].
There is considerable uncertainty in these observations given the different ways these data have been measured, standardized, and reported. For example, studies have expressed the number of amputations as a function of the total population (e.g., per 100,000 population), an at-risk population (e.g., per 10,000 people with diabetes), or as a true rate that accounts for the time people are at-risk (e.g., per 1000 person-years) [
12]. Such variation in the incidence rate data are further complicated depending on which amputation procedures are counted. For example, some studies exclude people with toe amputations [
13] and as such, likely underestimate the true incidence rate of PFA given that about 60% of PFA affect one or more toes [
3,
14].
While the effect of these sorts of variations in method design have been discussed in the literature [
12,
15,
16], the extent to which they actually explain variation in the incidence rates has not been scrutinized in the context of PFA and TTA.
A systematic review of recent epidemiological research would provide a means to make sense of the various ways these incidence rate data have been reported and where possible, synthesize these data to describe the incidence rate and understand how this has changed over time in people with PFA compared to TTA. Critical appraisal of the method design would help explain variation in the incidence rates between studies and help reconcile the seemingly disparate data reported in the literature.
A more informed understanding of the incidence rate and prevalence data are important to establish how incidence rates and prevalence may have changed over time so that we can plan for the specialist health care needs of those facing the prospect of, and living with, PFA.
Therefore, the aims of this systematic review were to:
(1)
describe the incidence rate and prevalence of dysvascular PFA,
(2)
describe whether the incidence rate and prevalence of dysvascular PFA has changed over time,
(3)
describe causes of variation in the incidence rate and prevalence reported,
(4)
compare the incidence rate and prevalence of dysvascular PFA and TTA.
Methods
Prior to conducting this review, a detailed systematic review protocol was registered in PROSPERO (CRD42015029186) and published [
17]; hence, a summary of the methods related to the epidemiological review have been reported here. We highlight that the protocol also included aims related to the outcomes of amputation which have been published in another systematic review [
1].
Search strategy
A search of the literature was systematically conducted using MEDLINE, EMBASE, psychINFO, AMED, CINAHL, ProQuest Nursing and Allied Health. Search terms related to the population and outcomes of interest were used in conjunction with wildcards and Boolean operators as part of a title, abstract, and keyword search [
17]. Each search strategy was developed, tested and refined by comparing the precision and comprehensiveness of the articles retrieved to a bank of known articles on the topic [
17].
All searches were limited to articles written in English given that such language restriction does not alter the outcome of systematic reviews and meta-analyses [
18,
19]. The search was restricted to publications since 1 January 2000 given that changes in treatment practices (e.g., common place use of revascularization prior to, or in conjunction with, amputation) have affected outcomes over time [
8,
10,
11].
Consistent with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, [
20] an illustrative search for one database is shown in Table
1.
Table 1
Example search for the CINAHL database to identify incidence and prevalence literature for people with dysvascular partial foot and transtibial amputation
1. | MH | “Amputation” |
2. | MH | “Amputees” |
3. | TI,AB,SU | (amput* AND (major OR lowerlimb* OR “lower limb”* OR “lower extremit*” OR “limb loss” OR LEA OR LLA)) |
4. | TI,AB,SU | (amput* AND (transtibial OR “trans tibial” OR belowknee OR “below knee” OR (below W2 knee) OR TTA OR BKA)) |
5. | TI,AB,SU | (amput* AND (minor OR “partial foot” OR Chopart* OR Lisfranc* OR tarsometatarsal OR transmetatarsal OR midtarsal OR “mid tarsal” OR midfoot OR “mid foot” OR ray OR phalangeal OR metatarsophalangeal OR toe* OR transtarsal OR “trans tarsal” OR TMT OR TMA OR MTP OR PFA)) |
6. | | 1 OR 2 OR 3 OR 4 OR 5 |
7. | TI,AB,SU | “incidence” OR “rate” OR “incidence rate” OR “prevalence” OR “trend” |
8 | | 6 AND 7 |
9. | | Limit 8 to English language |
10. | | Limit 9 to publication date: 01 January 2000 to 31 Decemeber 2015 |
11. | | Limit 10 to peer reviewed, academic journals |
Reference lists of included articles were hand searched to ensure that relevant publications were not missed. A forward-citation search using Google Scholar was conducted to identify early on-line articles published since the 1 January 2014 that had not yet been indexed in traditional databases [
21‐
23].
Data management
Search results from each database were exported into a shared EndNote X7.2.1 library (Thomson Reuters Inc., Philadelphia, PA, USA.) and duplicate records deleted [
17]. Full-text articles were retrieved and linked to the corresponding EndNote record. Bibliographic information were exported into an Excel 2013 (Microsoft Corporation Inc., North Ryde, Sydney, Australia) spreadsheet to track details about exclusion and full-text retrievals. The same spreadsheet was expanded for data extraction and critical appraisal [
17].
Selection process
Inclusion criteria were as follows:
1.
Peer reviewed studies of original research, irrespective of their design;
3.
Published between 1 January 2000–31 December 2015;
4.
Discrete cohort(s) with either: dysvascular PFA (irrespective of the level of PFA) or dysvascular PFA and TTA; and
5.
Reported data on the incidence rate or prevalence.
The International Standards Organization (ISO) definitions [
24] of TTA and PFA were used. As such, all levels of PFA (including amputation of one or more toes) were included but ankle disarticulation (i.e., Syme’s amputation) was excluded. Articles were included regardless of the way the numerator or the denominator were operationally defined [
17].
Search results were screened by one investigator based on review of the title, abstract, or full-text as necessary. After screening, all full-text articles were retrieved and reviewed independently by two investigators to confirm inclusion. The opinion of a third investigator was sought in cases of disagreement, and discussion occurred until consensus was achieved.
Quality appraisal/risk of bias in individual studies
Methodological quality and sources of bias were assessed using the McMaster Critical Review Forms [
25,
26] given this study formed part of a larger review into the outcomes of PFA and TTA [
1] that included studies of various designs [
27]. The McMaster Critical Review Forms include structured guidelines to reduce the likelihood of errors with their use [
28]. The quality appraisal was collated using Excel with detailed comments included to support the checklist items [
17].
Socio-demographic, methodological, results, and quality appraisal details were recorded in an Excel spreadsheet (Additional file
1) for each included article using the Cochrane Consumers and Communication Review Group’s data extraction template [
29]. The data extraction spreadsheet was piloted and refined prior to use [
17].
Two reviewers independently appraised included articles. Data were extracted from each article by a primary reviewer and checked by a second reviewer for accuracy and clarity. A third reviewer was called upon to appraise the article and contribute to the consensus decision as necessary. Authors of the original research were contacted for additional information or to clarify method details. Reminder emails were sent if a response was not received.
In cases where incidence rate or prevalence data were reported in figures only, authors of the original research were contacted to obtain these data. Where we were unable to obtain these data from authors, we utilized software (DigitizeIt v 2.2.2,
www.digitizeit.de) to digitize the figures and extract the
x,
y coordinates. This approach has been shown to be valid and reliable in several studies, and we adopted recommendations for minimizing errors, such as zooming in to identify the center of data points [
30‐
32].
For articles where data for the same participants were reported, subject numbers, demographics, and outcomes were compared across studies for discrepancies. Any uncertainty about the similarity in study participants and results, were clarified by contacting the authors of the original research. Where the same subjects were included in multiple studies, data were treated as a single source but all studies were cited.
Data summary and reporting
Extracted data were explored to identify variation in the way the prevalence or incidence rate data were reported and, where possible, efforts were made to reduce apparent variation. For example, studies that expressed the incidence rate per 10,000 or 100,000 people with diabetes were scaled to a common denominator (e.g., per 100,000 people with diabetes) to reduce variance and facilitate synthesis. Results were presented in separate sections for each denominator (e.g., per 100,000 people with diabetes) and included various subgroup analyses (e.g., stratified by diabetes type) with a view to synthesizing like data while preserving information inherent in the different strata reported. As part of the narrative review, issues with internal and external validity that most influenced the incidence rates or prevalence were discussed with a specific focus on limitations that lead to imprecision and heterogeneity [
33].
Where possible, descriptive statistics were used to summarize data. A mean annual incidence rate was calculated for each of the included studies; using the age- and sex-standardized incidence rates in preference to the crude incidence rates where possible. Notable variation between the crude or age- and sex-standardized incidence rates were highlighted through the narrative given the results were influenced by the population structure. To synthesize data across studies, the mean annual incidence rates were weighted by the average number of amputations each year to produce a point estimate and 95% confidence interval using StatsDirect3 (StatsDirect Limited, Cheshire, UK,
www.statsdirect.com).
Discussion
The purpose of the review was to develop an informed understanding of the incidence rate and prevalence of dysvascular PFA, how these compared to TTA, and how they have changed over time. To the best of our knowledge, this is the first systematic review that has endeavored to synthesize the incidence rate and prevalence data of any type of lower limb amputation. While systematic reviews of epidemiological data are common in many areas of healthcare, they have typically only been performed in well-developed bodies of literature where numerous studies report incidence rates using the same denominator, thereby facilitating synthesis using meta-analysis. We suggest that systematic reviews of epidemiological data in lower limb amputation have not previously been undertaken given the wide variety of methodological and reporting approaches that make the literature too heterogeneous to synthesize using statistical approaches. However, this does not negate the need for a clear understanding about the incidence rate or prevalence and how these have changed over time.
As highlighted in this review, the many different methodological and reporting approaches added considerably to the challenge of synthesizing data. We struggled to bring together data from such disparate studies until we recognized that we could reduce much of the apparent variation between studies by standardizing the incidence rates to common denominators; something only evident to us after careful critique of the way the denominators were operationally defined. By standardizing the incidence rates to common denominators where appropriate, we were able to reduce the apparent variation between studies that, in turn, made it possible to glean new insights into these studies leading to a more informed understanding.
In contrast to our initial impression of the literature, we were surprised by how homogenous the incidence rate were once studies were appropriately grouped by the same denominator and strata. We do not imply that there is absolute agreement between studies, just less variation than might be expected based on a primafacie evaluation of the epidemiological literature.
There were examples where the incidence rates of PFA were very homogenous or where variation between studies did not reduce confidence in the conclusions. For example, studies that reported the incidence rate of PFA per 100,000 general population had a very narrow confidence interval (4.0 per 100,000 general population, 95% CI, 3.82 to 4.17). While there was less precision in the incidence rate of PFA for people with diabetes (94.24 per 100,000 people with diabetes; 95% CI 55.50 to 133.00) compared to those without (3.80 per 100,000 people without diabetes; 95% CI 1.43 to 6.16), it probably matters little to the conclusion that risk of PFA is significantly greater (about 25 times greater) for people with diabetes [
7,
9,
50].
In terms of diabetes type, the higher incidence rates typically associated with type 2 diabetes [
42,
47] may be confounded by inclusion of people with first and recurrent amputation in the same cohort. When stratified by both diabetes type and initial/recurrent amputation, incidence rates of initial PFA were similar in cohorts with types 1 and 2 diabetes [
38]. In comparison, the incidence rate of recurrent PFA was 23 times higher in people with type 1 diabetes and 100 times higher in people with type 2 diabetes [
38]; highlighting the very high risk of recurrent PFA, particularly in people with type 2 diabetes.
The inclusion or exclusion of people with amputation of the toe(s) had a profound effect on the incidence rate of the PFA cohort, which may not be suprising given the majority of PFA affects one or more toes. By comparison, the inclusion or exclusion of people with more proximal level of PFA (e.g., tarsometatarsal or transtarsal amputation) did not seem to have a dramatic effect on the incidence rates.
When people with toe amputation were included in the PFA cohort, the incidence rates were comparable to TTA [
48,
49]. Given the time period during which these data were collected, there is uncertainty about the generalizability of these results to more contemporary healthcare settings.
There is little certainty in whether the incidence rates of PFA have remained stable over time or changed. To some extent, the uncertainty reflects the small number of like studies when considered with respect to the different denominators and strata. A number of common method design issues further complicated our understanding. Time series were often too short for small changes in the annual incidence rate of PFA—typically less than 1–2% per annum—to become sufficiently large to be statistically significant. More dramatic reductions in the annual incidence rate of PFA reported in the literature should be interpreted with caution given they were able to be explained by common method design issues such as the difficulties in accurately estimating diabetes prevalence or inappropriate inferential analyses. Studies with small subject numbers were suseptible to chance variation in the number of cases in any given year and, as such, the year-to-year variability tended to dwarf any small, cumulative, change in the annual incidence rate. While most studies used inferential analysis techniques designed to test for trends in epidemiological data, there was often little consideration about the suitability of the statistical tests given the descriptive data presented. For example, the chi-square test for trend can exaggerate the significance of the change over time where the annual incidence rate were not linear. Given the uncertainty introduced by these method design issues, it is unclear whether the incidence rate of PFA has changed over time or how this compares to changes over time in those with TTA.
As the first systematic review to synthesize epidemiological data describing the incidence rate and prevalence of PFA, we have been able to benchmark what could be considered typical incidence rates for each of the common denominators and strata. These data may be particularly valuable where sufficient studies made it possible to calculate a point estimate and 95% CI.
Future research
Given the insights gleaned from this review, we suggest that there is little opportunity to extend our understanding based on epidemiological studies of isolated hospitals, short-time series, and small numbers of amputations per annum. Large-scale epidemiological studies over lengthy time series are required. In all likelihood, these studies will require linked datasets of state or national amputation surgeries that include the thousands of people per annum needed to stratify by important risk factors. Only in this way, can we corroborate insights about the influence of diabetes type, initial/recurrent amputation or race, and thereby clarify our understanding.
In the same vein, studies that test for trends over time should ensure that the assumptions of the inferential analysis match the descriptive data and consider more sophisticated approaches, such as joinpoint regression, that may better test changes in trends within the time series. More contemporary prevalence estimate are also desperately needed, as are studies that stratify by race.
Limitations
Given the myriad of ways the incidence rates were reported in the literature, we made a number of pragmatic decisions to be able to synthesize the results across like studies. We collapsed crude or age- and sex-standardized incidence rate data given that in the two studies [
9,
37] that reported both these data, the incidence rates were highly correlated (
r > 0.98), suggesting that, for these two studies, and probably others, combining crude and standardized incidence rates would be unlikely to change our observations.
We standardized like denominators (e.g., incidence rate per 10,000 or per 100,000 general population) to reduce apparent variation between studies. Where we have done so, we have made it clear in the results narrative given the absolute incidence rates reported in the review will differ from those reported in the original research. We argue that this approach reduced apparent variation that made it possible to see the similarity in incidence rates across studies and identify true, not apparent, sources of variation.
We felt the only way to present these incidence rates were by subgroup within each denominator and thereby preserve information inherent in the different strata. Given the number of subgroup analyses, and the detailed narratives contextualizing the risk of bias, we did not feel it was necessary to include the risk of bias tables within the body of the manuscript; particularly given the McMaster Critical Review forms included a written appraisal of each article to supplement the check list items. We have reported the complete risk of bias assessment for each article as part of the appendix for readers wanting this level of detail (Additional file
1).