RESEARCH DESIGN AND METHODS

Patient population

Since 1988, Curative Health Services (CHS) has managed >150 distinct wound care facilities in 38 states in the U.S. and has maintained an administrative and patient record database (17). We have previously demonstrated via chart review the validity of using this database to study individuals with DNFUs (17,18). Briefly, in that study, we showed that we could use the database to accurately determine whether a subject had a diagnosis of DNFU and healed by the 20th week of care compared with the medical record (positive predictive values of 98% [95% CI 0.89–0.99] and 93% [0.68–0.99], respectively) (18). In the medical record, arterial flow abnormalities were determined noninvasively by TcPo₂ monitoring or arterial Doppler and neuropathy by the absence of a response to a 10-g Semmes-Weinstein filament. By using this algorithm, we specifically did not study individuals with significant lower-limb arterial disease.

Subjects for this investigation had been treated at a CHS center between 1988 and 2000 and had at least one DNFU. We analyzed the subject’s only or most recent registration period. To clarify, for a subject who was successfully treated, stopped receiving care at a CHS center, and then returned for care (i.e., re-registered), only the last cycle of wound care was analyzed. Furthermore, to avoid including subjects who were one-time specialty center consultations, any individual who did not have a second office visit or documentation of a surgical procedure within 6 weeks of the first office visit was excluded.

Outcome and risk factors

For all subjects, the outcome—a healed wound by the 20th week of care—was determined using a previously validated algorithm (18). We selected this outcome because this is the longest end point consistently used in clinical trials. We also evaluated a 12-week end point. The results from this analysis were similar to those reported below.

For this study, a risk factor was a subject or wound characteristic noted at the first office visit that might play a role in causing or explaining the outcome. The following risk factor variables were evaluated: patient age, patient sex, duration of the wound (months), size of the wound (mm²), wound grade (Table 1), number of wounds, prior care at a CHS center, and which CHS center. Prior care was defined by determining if the patient had previously received a full cycle of care (i.e., they previously had been registered by, treated by, and discharged from active care at a wound care center).

Analysis

All risk factor variables were characterized by estimating means, SDs, medians, and quartiles and by a visual assessment with a particular focus on skewing. Two variables in this data set, duration of the wound and wound size, were severely skewed (see results). These variables were log transformed to a more symmetric distribution. For the initial analyses, we used univariate assessments of association, such as 2 × 2 tables, χ² statistics, Mantel-Haenszel statistics, and one-way ANOVA.

Our primary question was to estimate the association of various risk factors with the probability that a patient with a DNFU would heal by the 20th week of care. To assess the magnitude of the effect of a given risk factor, we used single-variable (unadjusted) and multiple-variable (fully adjusted) logistic regression models to estimate odds ratios (ORs). Both unadjusted and fully adjusted ORs are reported with 95% CIs. Fully adjusted ORs were calculated by including all risk factor variables. Two-way interaction terms were evaluated. Both fixed effects and random effects models using generalized estimating equations were calculated (19). Finally, to simultaneously evaluate the effects of clustering due to wound care center and patient, we evaluated effect estimates using a generalized linear latent and mixed model (GLLAMM) (20). These techniques adjust the variance estimates because multiple observations may have been made within nonindependent units, such as a subject or wound care center. All analyses were conducted using SAS version 8 or STATA 7, or both, for a PC.

RESULTS

The 31,106 individuals who met our criteria for a DNFU had a total of 72,525 wounds (Table 2). Of the patients, 53.9% were men, and 20.5% had previously received care at a wound care center. The mean patient age was 63.8 years. Of the wounds, 76.2% were grade 2 or lower. Before log transformation, the mean duration of a wound was 5.39 months and the median duration of a wound was 1.0 months. The mean on the log scale was 0.48 log months (geometric mean 1.6 months), SD 1.39. Before log transformation, the mean size of a wound at presentation was 588.6 mm² and the median size of a wound was 118.0 mm². The mean on the log scale was 4.86 log mm² (geometric mean 196 mm²), SD 1.68. By the 20th week of care, 50.3% of the wounds healed. The percentages of wounds not healed for each risk factor are listed in Table 2.

Men were slightly less likely to heal (Table 3). Wounds on individuals with increasing numbers of wounds, wounds that were larger (log mm²), and wounds that were older (log months) were all less likely to heal. Wounds of increasing grade were also less likely to heal. We dichotomized wound grade, as presented in Table 1, as a grade of ≤2 vs. >2. This dichotomization is consistent with most randomized clinical trials in that they generally included only patients of grade 2 or lower (8,10,12,21). Unexpectedly, those who had previous wounds were more likely to heal. With the exception of wound grade, the unadjusted versus fully adjusted effect estimates did not differ by >10%, indicating that the unadjusted estimates were not confounded by the other risk factor variables (22). Several two-way interactions were evaluated: wound grade and duration, wound grade and area, wound grade and wound count, and duration and area. Whereas many of these interactions had P values of <0.01, the clinical significance of these interactions was less clear because the effect estimates for any specific wound grade varied very little compared with the fully adjusted point estimates (Table 2). For example, the largest change in the effect estimate of wound log duration occurred between wound grades 1 and 5 (OR 1.25 [95% CI 1.14–1.24] and 1.13 [1.05–1.22], respectively), and the largest change for wound log area occurred between grades 1 and 5 (1.19 [1.14–1.24] and 1.34 [1.26–1.42], respectively).

Some patients received an adjuvant treatment in association with standard therapy. Entering adjuvant therapies (platelet releasate, recombinant human platelet-derived growth factor, or graft skin) into our multivariate model did not change the point estimates of any of our risk factor variables by >10% (22). The results also did not differ if these patients were dropped from the analysis.

Because multiple patients may have been treated at the same wound care center and more than one wound per patient may have been used in our investigation, the data in our investigation may lack true independence. To evaluate this effect with respect to wound care center, we first entered each center, as a categorical variable, into a logistic regression model. Patient care by a specific center, compared with any other center, was not associated with wound healing. Next, we added each center as a categorical variable into the multivariate model containing all of our risk factors. This adjustment did not alter any of the fully adjusted effect estimates reported in Table 2 by >10% (22). Finally, we ran generalized estimation equations (GEEs) to compensate for the lack of independence due to wound care center, patient, and wound care center and patient simultaneously. We also ran this model clustering on patient and a GLLAMM model clustering on both patient and hospital. The GEE- and GLLAMM-estimated confidence intervals did not vary substantially from those presented in Table 3. Finally, it should be noted that for most clinical trials only one wound is enrolled and evaluated. To determine whether the effect estimates for our risk factor variables might be different for a single wound than for all of the wounds, we repeated our analyses only including the first or primary wound on an individual cared for in a wound care center. As can be noted in Table 3, the effect estimates for a single wound are not different by >10% from the effect estimates for all wounds.

CONCLUSIONS

Multiple clinically relevant risk factors exist that can be used to distinguish between a wound that is likely to heal and one that is not likely to heal by the 20th week of standard therapy for a patient with a DNFU. The risk factors or wound characteristics that most dramatically are associated with a wound failing to heal are increasing wound size, increasing wound duration, and the grade of the wound. More simply, wound grade can be redefined such that wounds are differentiated by whether they penetrate through the subcutaneous fat layer (i.e., grade 3 or higher). It is these that are less likely to heal. These risk factor variables are clinically relevant enough that they should be evaluated during the first office visit by the physician or health care provider caring for a patient with a DNFU.

The association between two risk factor variables and wound healing may be confusing. These risk factors are prior wounds (patients with prior wounds were more likely to heal) and the use of adjuvant therapies (those that used adjuvants were less likely to heal). We believe that the explanations for these associations are similar and related to selection bias. We hypothesize that those who had prior wounds were more likely to seek care at the wound care center that successfully treated them in the past. In other words, those who successfully responded to the wound care center environment were likely to return and likely to be successful again. With respect to use of a treatment adjuvant, these therapies did not start at the first office visit, so it is unlikely that a patient received a full 20 weeks of care with adjuvants (i.e., our evaluation of these therapies is incomplete). Second, because these therapies often did not commence until week 8 or 12, it is very likely that physicians selected those patients that appeared to be doing poorly for adjuvant care. In fact, in a previous publication in which we evaluated the use of platelet releasate using a statistical technique that models selection bias, we were able to show that the most severe wounds received this adjuvant treatment and that it was effective (17).

Recently, Oyibo et al. (23) evaluated the association of several risk factors, including wound size and measured wound depth (similar to our wound grade), with healing in patients with diabetes and foot ulcers. Their cohort study included 194 patients evaluated in a foot ulcer center in England and a center in Texas. As in our study, they noted that the size of the wound correlated with the time required for a wound to heal. Unlike our study, the tissue depth of the wound (our wound grade) was not predictive of time to healing but did predict amputation risk. Many other risk factors did not correlate with healing, including age and sex of the patient, type of diabetes, duration of diabetes, and ulcer site. Their study differed from ours in that they evaluated only one wound per patient; our patients on average were older, more of our patients were female, all of our patients were neuropathic with good arterial flow, we evaluated healing by the 20th week of care and present the odds of healing, and our study was larger and multicentered.

In fact, one of the greatest strengths of our study is its sheer size and the geographic distribution of wound care centers throughout the U.S. Because our data source was CHS wound care centers, it is possible (but we believe unlikely) that the results of this study do not generalize to independent practitioners or those who do not frequently treat individuals with DNFUs. It is also important to note that wound care center was not a significant confounding variable. In other words, wound duration, wound size, and wound grade had similar effects in all centers. Furthermore, adding wound care center did not significantly improve our models. It seems likely that because CHS requests that local wound care centers follow their protocols—which include discussions about debridement, good wound care, offloading, etc.—it is possible to successfully educationally intervene across a large network of wound care specialists. Finally, it should be noted that the association of wound size and wound duration with wound healing has been noted in meta-analyses of clinical treatment trials for DNFUs and venous leg ulcers, another chronic wound (15,24,25).

There are several limitations to this study that are common to observational database studies. We discussed selection bias above. Information bias may have also affected our results. First, there may have been systematic differences between the reporting of those who did and did not have particular risk factors. Based on the risk factors that we evaluated, this is unlikely—our risk factors were basic to the initial evaluation of a chronic wound. Second, it is possible that there were systematic differences between the way risk factors were measured and the likelihood that a wound would heal. For example, it is possible that during the first patient visit the physician undermeasured wounds that were likely to heal and overmeasured wounds that were unlikely to heal. We believe that this is unlikely because wound size is also used to monitor the progress of a wound. In addition, those recording the data were not aware of our study. Finally, as with all observational studies, the evaluation of risk factors and confounding factors is limited to variables accurately and consistently present in a database. For this study, we were not able to evaluate glycosylated hemoglobin (which was previously shown not to influence wound healing [24]), fasting blood glucose, presence of retinopathy or nephropathy, body weight, location of the ulcer, degree of foot deformity, or the use of or compliance with treatments used for diabetes.

In summary, we have shown that at the initial visit, the size of the wound, the age of the wound, and the grade of the wound are all predictors for the failure of a neuropathic foot ulcer to heal. These are simple risk factors that are easily measured at the initial office visit. In most cases, they are already part of the initial patient assessment. We believe that, taken together, these risk factors argue for the early treatment of individuals with foot ulcers. The association of these risk factors with a wound healing is the same whether the wound receives standard care or standard care plus an adjuvant (such as a topical growth factor). More importantly, we have demonstrated that these associations depend on the wound and that they do not vary from wound care center to wound care center. Finally, our results are important to clinicians caring for patients with DNFUs in that it is possible to characterize those wounds that will do well with medical therapy and those that will not. In addition, for those designing trials, we believe that it is prudent that these risk factors be considered both when choosing whom to accept into a trial and when assessing the effect of a novel therapeutic.