Causation Research on Diabetic Foot Complications–What I Learned From Roger Pecoraro: The 2021 Roger E. Pecoraro Award Lecture

Who Was Roger Pecoraro?

Roger was born in 1944 in Elmira, New York. His family moved to Arizona in 1957 due to a health condition in a family member. He graduated summa cum laude from the University of Arizona in 1965 and received his MD degree from the University of Washington School of Medicine in 1970. He completed his internship and residency in internal medicine at UCLA-LA County Harbor General Hospital in 1973 and next completed fellowship training in Endocrinology at the University of Washington between 1973 and 1975. He joined the Department of Medicine at the University of Washington in 1975 as a faculty member and held the rank of Associate Professor at the time of his death from cancer at age 48 years on 17 December 1991 (1). He is survived by his wife Janice and sons Jesse and Shannon. As no photos of Roger are available online to my knowledge, I have included one from an unknown date, but my best estimate is that it was probably taken 2–3 years prior to his death (Fig. 1).

Figure 1

Roger E. Pecoraro, MD. Undated photo.

Roger’s Influence on My Career Path in Diabetes Research

I first met Roger in 1982 as a Robert Wood Johnson Clinical Scholars Fellow at VA Puget Sound in Seattle. I accepted a position as Assistant Professor at the University of Washington, moving there from a faculty position at the University of Colorado in 1989. Roger contacted me prior to my arrival in Seattle to recruit me to collaborate with him on his research. Between 1989 and 1991, we worked together on the design and initiation of the Seattle Diabetic Foot Study, including obtaining the first successful funding for this project in 1991 from the Department of Veterans Affairs (VA) Rehabilitation Research and Development (RR&D) Service.

Roger held several beliefs regarding diabetic foot complications based on data from his own research and that of others and his own opinions. He did not accept that diabetic foot complications were inevitable and believed that the development of a diabetic foot ulcer was the key in the cascade of events leading to diabetic limb amputation. One of his favorite articles was on the general subject of causation by Kenneth Rothman published in 1976 (2). Rothman proposed that what he refers to as a “sufficient cause” is necessary for disease causation and that it inevitably leads to disease or an adverse outcome such as limb amputation. He further argues that a “sufficient cause” is comprised of several “component causes.” This is illustrated in a figure that Roger reproduced in an article he published in 1990 (3) (Fig. 2). Shown in this figure are five component causes labeled A through E. Each of these component causes is not sufficient to cause a disease. All five component causes are required to cause the disease, and when all five are present they comprise a sufficient cause that inevitably produces an effect. The important implication of this model of causation is that removal of just one component cause prevents disease, even though the four other component causes remain.

Figure 2

The pie diagram shown to the left represents the sufficient and component causal framework. A–E represent causes that are not sufficient in themselves but that are required components of a sufficient cause that inevitably produces an effect. The right side represents the causal pathway to amputation, which includes essential contributions from the component causes shown, all of which comprise a sufficient cause. Theoretically, amputation could have been avoided by elimination of any one component cause. Adapted from Pecoraro et al. (3).

The influence that this model of causation had on Roger’s thinking is clearly evident from his 1990 article on pathways to diabetic limb amputation (3). Coauthors with him on this article were Gayle Reiber, an epidemiologist and past recipient of the Roger E. Pecoraro Award, and Ernie Burgess, an orthopedic surgeon renowned for his contributions to surgical amputation methods and lower limb prosthesis design innovations. As seen in the figure appearing in the article and reproduced here in Fig. 2, Roger applied the Rothman causal model to the problem of diabetic limb amputation, demonstrating five component causes (neuropathy, minor trauma, ulceration, faulty healing, gangrene) leading to a sufficient cause that results in amputation. Roger further defined seven component causes that he believed were involved in the causation of diabetic lower limb amputation: ischemia, neuropathy, trauma, ulceration, faulty wound healing, infection, and gangrene. He proceeded to describe the causal pathways to lower limb amputation in a case series of 80 patients with diabetes and limb loss. The most common pathways are shown in Supplementary Fig. 1, reproduced from his 1990 article (3). The most common sufficient causes, occurring in 25% of amputations, included ischemia, neuropathy, trauma, ulceration, and faulty wound healing. He noted the consistent occurrence of two component causes in the majority of sufficient causes, these being ulceration and faulty wound healing. This research shaped his thinking by confirming that the two opportunities for intervention to avoid limb amputation were prevention of ulceration and timely healing of ulcers. It laid the foundation for the Seattle Diabetic Foot Study. Roger believed that a prospective study of foot ulcer that investigated many potential factors would improve our understanding of the component causes of diabetic foot ulceration and create possibilities for prevention through elimination or amelioration of such causes.

Key Findings of the Seattle Diabetic Foot Study

Recruitment of participants for the Seattle Diabetic Foot Study began in 1990. It was designed as a prospective study of general internal medicine patients with diabetes and without foot ulcer to identify risk factors for the development of foot ulceration and limb amputation over time, with the expectation that at least 5 years of follow-up would be needed to achieve its objectives. As Roger died ∼1 year into the planned follow-up, I took on the responsibility of leading this research effort. As it turned out, additional follow-up was needed to observe enough end points to achieve our research objectives and also to incorporate new measurements of interest that became available during follow-up, such as in-shoe plantar pressure measurement. The odds were certainly stacked against success from the onset of the Seattle Diabetic Foot Study due to the daunting task of recruiting and following over time the hundreds of participants needed to achieve our research goals and the very limited duration of funding available at the time to support our research. VA RR&D funding at the time was limited to 3 years. Funding for this research ceased in 2002 after 12 years, for a total of four grants awarded to support it.

In the Seattle Diabetic Foot Study we conducted an extensive evaluation at the two-part baseline visit, which included medical history and physical examination that focused on the lower limbs, assessment of functional status, and multiple neurovascular measurements including ankle-brachial index (ABI), hallux blood pressure, transcutaneous oximetry (TcPO2) to assess skin oxygenation in mmHg, sensory neuropathy by 10-g monofilament testing, autonomic function assessed by cardiovascular reflexes (R-R variation with timed inspiration, orthostatic blood pressure drop), and laser Doppler flowmetry (4). The baseline evaluation was repeated where possible annually to assess outcome occurrences as well as change in measurements of interest. Follow-up was conducted by quarterly mailed letter assessing occurrence of foot outcomes of interest, and if a positive response was received the participant was evaluated by the research team to determine whether a foot complication had occurred.

The Seattle Diabetic Foot Study incorporated several strong methodologic features, but from the outset there were also limitations that we recognized but could not overcome at the time. Strengths included the prospective design, which permits assessment of temporal sequence to allow confirmation that a potential cause preceded an outcome and was not due to it. The results would be generalizable to nonspecialty populations of patients with diabetes as would be seen in primary care clinics. Extensive limb-specific neurovascular measurements were performed, which permitted analysis of foot ulcer and amputation risk factors by limb. Statistical analyses were planned to be performed with each limb as subject, avoiding the problem of choosing measurements from only one limb in person-level analyses, thereby creating potential misclassification due to the possibility of ulceration occurring on a limb the measurements for which were not included in the analysis. Several limitations were presented for which there was no solution at the time of study initiation. First and foremost, few women participants were available in the veteran population. Also, follow-up for ulcer occurrence terminated after the first incident ulcer because of the relative difficulty of using statistical methods with the ability to consider multiple outcome occurrences. Lastly, the limit on duration of funding and the need to repetitively obtain additional funding through competition every 3 years finally led to termination of funding before length of follow-up permitted a thorough study of amputation risk.

Available space does not permit discussion of all 22 articles that have been published based on the Seattle Diabetic Foot Study (4–25). I present summaries of selected articles based on this research that are most pertinent to its main objectives.

Risk Factors for Diabetic Foot Ulcer

The first publication on risk factors for diabetic foot ulcer with use of the prospective study design appeared in 1999 and included 749 patients, with 1,483 limbs, 98% of whom were men and 94% of whom had type 2 diabetes (4). The mean follow-up was 3.7 years, mean age was 63 years, and mean diabetes duration was 11 years. A total of 162 foot ulcers occurred, for an incidence of 3.0 per 100 limb-years. Later reports from this research included larger sample sizes, as enrollment was rolling and ongoing at the time of the analysis reported here. Of the 47 potential risk factors examined, 32 were significantly related to foot ulcer risk (P < 0.05) in univariate Cox models including greater height and weight; longer diabetes duration; insulin use; higher random plasma glucose and HbA_1c; higher erythrocyte sedimentation rate and serum creatinine; lower ankle blood pressure, ABI, and dorsal foot TcPO2; claudication with walking less than one block; self-reported clinician-diagnosed peripheral vascular disease; history of vascular bypass surgery; greater orthostatic blood pressure drop and sensory neuropathy assessed by monofilament testing or 128-Hz tuning fork; absent Achilles tendon reflexes; foot pain; foot numbness; clinician-diagnosed neuropathy; history of foot ulcer and lower limb amputation; wearing special footwear; hallux limitus; hammer/claw toe or Charcot deformity; lower hallux joint mobility; abnormal extensor digitorum brevis test; lower limb edema; history of laser photocoagulation treatment; and <20/40 vision. No significant association was seen between foot ulcer risk and the following characteristics: age, race, pack-years smoked, history of diabetes education, hallux blood pressure, laser Doppler flow on the dorsal foot, hallux TcPO2, prominent metatarsal heads, bony foot prominences, hallux valgus, ankle joint mobility, and self-reported physician-diagnosed nephropathy.

A stepwise multivariable Cox model was constructed that only included variables with P < 0.20 in univariate analysis. The results of this analysis identified nine factors significantly related to foot ulcer risk, as shown in the forest plot in Fig. 3. Factors independently associated with higher risk included sensory neuropathy by monofilament testing, history of foot ulcer or amputation, insulin use, greater body weight, vision <20/40, and orthostatic blood pressure drop, while independent predictors of lower risk were higher ABI and dorsal foot TcPO2. This result confirmed a multifactorial etiology for diabetic foot ulcer with identified factors reflecting dysfunction of autonomic and sensory nerves; poor perfusion; higher weight, perhaps reflecting greater plantar pressure or mobility limitations impairing ability to self-examine feet; poor vision, perhaps compromising self-care or signifying presence of diabetic retinopathy; insulin use indicating greater glycemia; and past history of prior diabetic foot complications. The identification of factors likely representing multiple component causes creates opportunities for preventive interventions, as reflected by American Diabetes Association 2021 Standards of Medical Care in Diabetes, which recommends more frequent foot examinations and referral to foot care specialists depending on presence of certain of these independent risk factors (26).

Figure 3

Multivariable Cox regression analysis of statistically significant (P < 0.05) risk factors for incident diabetic foot ulcer. Forest plot displays the point estimates of the hazard ratios as a square with the 95% CI as a horizontal line.

Prediction of Diabetic Foot Ulcer Occurrence Using Commonly Available Clinical Information

In the Seattle Diabetic Foot Study many measurements were performed that required specialized training or equipment that would only be available to a small minority of clinicians. This study did, however, include many readily available measurements that could be performed by any clinician. We next sought to develop a prediction model for diabetic foot ulcer occurrence using such readily available measurements (22). The measurements that we considered for this objective included demographic factors, diabetes characteristics, intermittent claudication, 10-g monofilament insensitivity, history of foot ulcer or amputation, foot deformity, callus, hallux limitus, pedal edema, tinea pedis, onychomycosis, poor vision, laser photocoagulation treatment history, and smoking. As our goal was to predict a person’s risk of ulceration, we performed a person-level rather than limb-level Cox regression analysis using a backward-selection algorithm. For limb-specific measurements, the worst of the two limb measurements was used in the analysis. A total of 1,285 patients were available, who developed 216 foot ulcers over an average follow-up of 3.38 years.

A total of seven factors emerged as significant independent predictors of ulcer, including sensory neuropathy, history of foot ulcer or amputation, higher HbA_1c, onychomycosis, and vision <20/40 (Fig. 4). Tinea pedis significantly predicted lower risk of ulcer occurrence. The AUROC for this model was 0.81 at 1 year and 0.76 at 5 years, showing good to excellent ability to discriminate by outcome. The risk score equation and instructions on how it is calculated are shown in Supplementary Fig. 2. The Kaplan-Meier curve shown in this same figure displays ulcer-free survival by risk score quartile. As can be seen in this figure, patients scoring in the fourth or highest quartile had <40% chance of remaining ulcer free at the end of follow-up.

Figure 4

Multivariable Cox regression model of statistically significant (P < 0.05) commonly available clinical risk factors for incidence of diabetic foot ulcer. Forest plot display as described for Fig. 3.

The finding of lower ulcer risk with presence of tinea pedis was not expected. We reasoned that since this superficial mycosis is thought to require perspiration to flourish, its presence might indicate better autonomic function. We therefore examined whether differences in autonomic function existed between patients with and without tinea pedis. Both greater heart rate variability with timed breathing (7.33 vs. 6.80 bpm, P = 0.020) and less orthostatic blood pressure drop (6.40 vs. 7.19 mmHg, P = 0.140) were seen in association with tinea pedis, arguing that its presence reflects better autonomic function, which evidence argues would lessen risk of foot ulceration.

We did not present a validation of this foot ulcer occurrence prediction model concurrent with or following its publication. However, unbeknownst to us, Monteiro-Soares and Dinis-Ribeiro conducted an independent evaluation among patients receiving care at a diabetic foot clinic at a tertiary hospital in Portugal (27). The study population included 360 patients, of whom 55% were women and 98% had type 2 diabetes. The AUROC of 0.83 of our model showed similar excellent ability in this population to discriminate between patients who did and did not develop foot ulcer over a median follow-up of 25 months. Most importantly, this validation showed that our model performed well in women, which we were not able to examine due to our predominantly male veteran population. Monteiro-Soares and Dinis-Ribeiro further refined our model by adding a variable to indicate use of high risk footwear (e.g., poor coverage of the foot such as sandals), which boosted the AUROC to 0.88 (27).

Risk Factors for Lower Limb Amputation

As mentioned earlier, in person evaluations of Seattle Diabetic Foot Study participants ended in 2002. Only 50 amputations had occurred during follow-up at that time, which limited analyses of risk factors for this outcome due to low power. To further capture this outcome, we searched electronic health records from the national VA corporate data warehouse and Centers for Medicare & Medicaid data for amputation ICD-9 procedure codes (84.1x) from 1 January 1990 to 31 December 2012 (25). Medical records of all participants identified as having an amputation from this search were reviewed to determine laterality and amputation level. A total of 136 amputations overall were identified, with 65% major, and yielding an overall incidence of 5.3 per 1,000 limb-years.

A multivariable Cox model was fit to the data with use of a backward selection algorithm while retaining all variables with P < 0.10 (Fig. 5). Factors associated with significantly higher risk included poor vision (20/70–20/200), lower estimated glomerular filtration rate, sensory neuropathy by 10-g monofilament testing, and ABI other than in the normal range of 0.91–1.29. Older age and higher body weight were both associated with lower amputation risk for unclear reasons. This finding for weight contrasts with our previous finding of higher weight associated with higher foot ulcer risk (4). A number of other factors, some of which were significant predictors of foot ulceration, were near misses with regard to showing a statistically significant association with amputation as seen in multiple CIs barely crossing the null value of 1.0 (Fig. 5).

Figure 5

Multivariable Cox model of limb- and person-level factors predicting amputation risk. Forest plot display as described for Fig. 3. Solid black lines indicate associations with 0.05 < P < 0.10. 44C, 44°C; BP, blood pressure; eGFR, estimated glomerular filtration rate; yrs, years.

Similar elevations in risk were seen with higher ABI (≥1.3) and the next to lowest category (ABI 0.51–0.90). High ABI results from incompressible vessels due to medial arterial calcification, but this finding does not necessarily reflect impaired perfusion. We examined whether other measurements of perfusion performed in our protocol could provide additional information about amputation risk in the high ABI subgroup. We examined dorsal foot TcPO2, which was not overall associated with amputation risk, but found that in this subgroup a value of <26 mmHg significantly (P < 0.05) predicted higher amputation risk. This result suggests a role for TcPO2 measurement in further defining amputation risk among persons with diabetes and elevated ABI.

Plantar Pressure and Diabetic Foot Ulcer Risk

Elevated plantar pressure is thought to be a key factor in the development of diabetic foot ulcer (28) and was reported to increase risk of ulceration in several prospective studies performed in persons initially without foot ulcer (29,30). However, this research did not examine whether pressure measurement at a certain foot location was related to risk of ulcer at that location. We performed F-scan in-shoe plantar pressure measurements during a 20-m walk in patients wearing their usual footwear (24). We divided the plantar foot surface into eight regions (24) (Supplementary Fig. 3). Peak plantar pressure measurements were available on 591 patients at each of the eight plantar locations, for a total of 8,880 measurements. The plantar locations of the 47 ulcers that occurred by the eight locations are represented in Fig. 6 as gray triangles overlaid on box plots depicting the median, 25th and 75th percentiles, range, and outliers for mean peak plantar pressure by foot location. The greatest number of ulcers occurred on the hallux, but at a similar number of pressures above and below the median, arguing against an association between pressure level and ulcer risk at this location. At the metatarsal head sites (MH1, MH2–4, and MH5), the majority of pressure measurements for incident ulcers are found above the median value with several >75th percentile and even two in the outlier range. A Cox model confirmed a significant increase in ulcer risk in comparison of the 75th percentile with the 25th percentile peak pressure measurement (hazard ratio 1.75, 95% CI 1.01–2.72). The third most frequent ulcer location site was the heel, which was similar to the hallux in demonstrating pressures associated with ulcer occurrence distributed above and below the median measurement. We concluded that greater mean peak plantar pressure is associated with higher foot ulcer risk over the metatarsal heads but that this relationship is dependent on foot location.

Figure 6

Box plots of mean peak pressure by plantar measurement location (n = 8,880) displaying the minimum, 25th percentile, median, 75th percentile, and maximum values at each location. Gray triangles indicate sites where ulcers developed and the mean peak pressure for that site in the subject who developed the foot ulcer. Lat MF, lateral midfoot; Med MF, medial midfoot; Toes, other toes. Reprinted with permission from Ledoux et al. (24).

Diabetic Foot Ulcer and Mortality Risk

At approximately the 4-year point in the Seattle Diabetic Foot Study, the study coordinator, Jessie Ahroni, noted what she believed was an elevated mortality in study participants who had developed a diabetic foot ulcer based on her clinical impression. Support for this impression came around that time with the halting by the U.S. Food and Drug Administration of a clinical trial of a copper peptide gel treatment for foot ulcer healing due to higher than expected mortality (31). The excess deaths were later ruled not due to the pharmaceutical. We decided to examine and quantify the risk of death in our cohort among patients who developed a diabetic foot ulcer during follow-up (8). Among persons who developed a foot ulcer, 11 deaths occurred during 91 person-years of follow-up, while 61 deaths occurred among the persons not developing a foot ulcer and who had accumulated 1,202 person-years of follow-up. The hazard ratio for mortality in a Cox model with treatment of ulcer occurrence as a time-dependent covariate was 2.38 (95% CI 1.22–4.63), indicating that the occurrence of a foot ulcer is associated with a greater than twofold subsequent increase in mortality risk in comparison with patients with diabetes who had not developed a foot ulcer. We believe that ours was the first report to quantify risk of mortality following incident foot ulcer occurrence.

What I Learned From Roger Pecoraro

Roger taught me several valuable lessons. He was not willing to accept a prevalent complacent attitude 30–40 years ago that diabetes complications were inevitable. He was open to new ideas and problem-solving approaches, such as causal models of disease, and he was guided by these ideas in conducting research that developed and answered important questions. He also taught me to pursue a valuable idea even if the odds of failure were high, as was truly the case at the beginnings of the Seattle Diabetic Foot Study. He was primarily responsible for instilling in me an interest in the study of diabetic foot complications, which I have pursued for the past 30 years.

Summary and Conclusions

The Seattle Diabetic Foot Study, which I led as Principal Investigator following Roger’s death in 1991, has made multiple contributions to our understanding of the etiology and consequences of diabetic foot complications. The research has identified multiple risk factors for diabetic foot ulcer and amputation that include not only neurovascular disease capable of leading to unnoticed trauma or tissue loss due to ischemia and past history of foot ulcer and amputation but also poor vision, body weight, insulin use, and chronic kidney disease that may be markers for disease severity or impediments to effective self-care. The multiple independent risk factor “component causes” identified support the causal model of disease causation favored by Roger. Additional contributions include a validated and accurate prediction model for foot ulcer development based on readily available clinical information, which potentially can assist in targeting of patients with diabetes in need of higher intensity of foot care services. Although abnormal plantar pressure is considered a key component cause of foot ulcer, our research suggests that this may only apply to the metatarsal head locations. Our demonstration of higher all-cause mortality among individuals who develop a diabetic foot ulcer argues that this foot lesion is a marker for serious disease elsewhere limiting survival.

Much has been learned about diabetic foot ulceration and amputation from the Seattle Diabetic Foot Study and other important research conducted by many investigators nationally and internationally on the problem of diabetic foot complications. Much more, however, remains to be done to prevent these complications. Despite a fall in diabetic lower limb amputations that began in the 1990s, recently the trend reversed, with an increase noted since 2009 in the U.S. mainly in men and specifically for toe amputations (32). Perhaps this is due to complacency and the expectation that the downward trend would continue. Although we have a much better understanding of risk factors for diabetic foot complications than we did 30 years ago, our knowledge of proven methods for prevention is still lacking, according to a recent systematic review of interventions to prevent a first foot ulcer (33). It is here that research seems likely to benefit our patients with diabetes and hopefully reverse the upward trend in nontraumatic diabetic amputations that began in 2009.