This is an observational cohort study with data from the Danish Nationwide Health registers. The RECORD guidelines for reporting routinely collected observational data were followed [11].

Data on all first-time MLEA procedures from January 1, 2010, to December 31, 2021, was obtained from the Danish National Patient Register (DNPR). DNPR contains data on all Danish hospitalizations, including ICD-10 diagnoses and NOMESCO surgical procedures since 1977, with an estimated completeness of > 99% [12]. Identification of cases and linkage between registers was possible with the unique social security number, a 10-digit personal number assigned to all persons living in Denmark [13]. Data from the DNPR include information on gender, amputation procedure, and ICD-10 diagnosis codes.

Data from The Danish National Prescription Database (DNPD) was used to enhance the validity of diabetes, hypertension and dyslipidemia diagnosis [8, 14]. The DNPD contains data on all reimbursed prescriptions from 1995 classified by Anatomical Therapeutic Chemical (ATC) codes [15].

The population includes all patients ≥ 18 years old undergoing a first-time primary MLEA procedure in Denmark from January 1, 2010, to December 31, 2021, defined through procedure codes (amputation on thigh/hip (KNFQ*) and amputation on lower leg/knee (KNGQ*) the DNPR (Fig. 1). To ensure that the index MLEA was a first-time MLEA conducted in the study period (2010–2021), all patients with an amputation procedure code; amputation on thigh/hip (KNFQ*) and/or amputation on lower leg/knee (KNGQ*) from 1996 to 2009 was excluded from the raw data set prior our obtainment (washout). Due to applied data protection rules in Denmark, The Danish Health Data Authority conducted this exclusion of prior cases. An overview of included cases is presented in Fig. 1.

Studies based on the same raw data, but with different study populations and aims, has been conducted [16, 17].

Hip disarticulation, transfemoral amputations, knee disarticulations, and transtibial amputations are classified as Major Lower Extremity Amputations (MLEA). An amputation through the ankle joint (Syme’s procedure) was not included as a major amputation, but as a minor amputation. Below/through-knee amputation (BKA) consists of transtibial amputation(TTA) and knee disarticulation(KD), and above-knee amputation (AKA) consists of hip disarticulation (HD) and transfemoral amputation (TFA).

Only first-time MLEAs are included as index procedures for analysis, defined as the first MLEA that occur for the patient. Revisions or contralateral MLEA are not considered first-time MLEAs.

Charlson Comorbidity index (CCI) was calculated based on data from DNPR. Diagnosis codes in a 10-year period prior to the index amputation were used and weighted to provide the CCI as described by Quan et al. [18].

Marital status was obtained from the Danish Civil Registration System [13].

The definition of the comorbidity variables for diabetes, cardiovascular disease(CVD), renal insufficiency, peripheral arterial disease (PAD), hypertension and dyslipidemia were based on ICD10 codes from DNPR, and for some diagnoses (diabetes, hypertension, dyslipidemia) also ATC codes from DNPD. As an example a patient was categorized with diabetes if one of the following ICD10 codes was registered: E10*, E11*, E13*, E14* or the patient had redeemed two or more anti-diabetic medicine prescriptions with the following ATC codes within five years before the index date: A10A (Insulins and analogs) A10B (Blood Glucose lowering drugs, excl. insulins), A10X (Other drugs used in diabetes). We used the same definision of a prior would as Madsen et al., but it dit not have to be considered chronic [19]. A complete overview of the definition of diagnoses and procedures is provided in the Appendix 1- Supplementary information.

– Flowchart of included cases. DNPR: Danish National patient registry

Descriptive statistics of baseline characteristics for each surgical strategy were performed. First-time MLEA rates per 100.000 inhabitants ≥ 18 years were calculated for each year in the study period using the annual number of procedures and the population ≥ 18 years. Incidence rates were estimated for each year and divided in gender and age group (< 50, 50–70, 71–80, 81–90, and > 90), and presented with 95% confidence intervals (CI). To calculate the annual incidence rates per 100,000 inhabitants in the beginning of each year, numbers from Statistics Denmark were used as the denominator, representing the population size at a given year, stratified in gender and age group. Poisson regression was applied to investigate a possible time trend adjusting for gender and age group. The investigated comorbidities were plotted over the study period. Additionally, we created the same plot, stratifying by initial above-knee and below-knee amputations (see supplemental Fig. 1A-1B).

Post hoc analysis of the data includes a sensitivity analysis of the incidence rates, where the denominator was adjusted for those who already had an amputation, as well as logistic regression analysis to explore significant statistical changes over time for the different surgical strategies and comorbidities during the study period. For all time trend analysis the year of surgery was treated as a continuous variable. Changes in the initial level during the study period were assessed individually for each level. We adjusted for comorbidities (diabetes, PAD, CVD, renal insufficiency, hypertension, dyslipidemia, and prior wound), prior surgery (vascular surgery and minor amputations), gender, CCI-group, and age group. The goodness of fit for the logistic models was checked, and found insignificant for TFA and TTA, but significant for KD and HD, properly due to a small number of observations in these groups. We investigated time trends for the comorbidities and prior surgery: diabetes, PAD, CVD, renal insufficiency, hypertension, prior wound dyslipidemia, prior vascular surgery, and prior minor amputation. All analysis were adjusted for gender, initial index level, age group, and CCI-group, in addition to all comorbidities and prior surgery except the one being analysed. The Goodness of fit for each variable was checked and found to be acceptable. The time trend for a high CCI-score, defined as ≥ 3, high age, defined as ≥ 70 years was analysed using a logistic regression model. This model was adjusted for all comorbidity and prior surgery variables, gender, initial index level, age group, and CCI group. However, the high CCI score was not adjusted for CCI group, and high age was not adjusted for age group. Statistical analyses were performed using STATA version 17.0.