The development of pressure ulcers (PrUs) remains a significant health problem, affecting more than 3 million American adults annually.1 Whittington and Briones2 examined hospital-acquired PrU prevalence rates over a 6-year sequential period and found rates between 7% and 9%. In 2004, 455,000 PrUs were identified in hospitalized patients.3 PrUs are also a costly condition, with the average charge being $43,180 per hospital stay.4 One analysis found that the cost for treating an infected Stage IV PrU in an intensive care unit can cost more than $120,000 when all costs are taken into account (eg, antibiotics, nursing time, medical time, etc).5 The Centers for Medicare & Medicaid Services (CMS) estimated spending in excess of $11 billion annually on reimbursing hospitals for PrU care.4 PrUs have also been associated with mortality. One study calculated the mortality rates associated with PrUs to be as high as 3.74 per 100,000 population.6
Preventing PrUs has been a prominent concern in healthcare as early as 1859, when Florence Nightingale (the mother of American nursing) noted that "if [the patient] has a bedsore, it is generally not the fault of the disease, but of the nursing."7 Although Nightingale's comments focus on nursing's responsibility, more current thinking suggests that the prevention of PrUs is a multidisciplinary responsibility.8-11 The US Agency for Healthcare Research and Quality (formerly the Agency for Health Care Policy and Research [AHCPR]) published national clinical practice guidelines on preventing PrUs in 1992.12 More recently, in 2009, the National Pressure Ulcer Advisory Panel (NPUAP) and European Pressure Ulcer Advisory Panel (EPUAP) published new joint prevention and treatment guidelines.13 Although the new NPUAP-EPUAP PrU prevention and treatment guidelines reaffirm many of the AHCPR recommendations, the 2009 version is substantially more complete and includes several entirely new areas.
Hospitals remain under increased scrutiny to implement comprehensive prevention strategies. Litigation remains rampant in hospitals-legal awards against hospitals as high as $85 million have been noted.14 On October 1, 2008, the CMS implemented a revolutionary program that would no longer reimburse acute care facilities for the development of a hospital-acquired Stage III or Stage IV PrU at a higher diagnosis-related group rate when these PrUs are billed as a secondary diagnosis.15 This increases the clinician's responsibility to detect PrUs present on admission (POA).
It is evident that the new regulation for nonpayment for hospital-acquired PrUs is leading to an increase in the number and comprehensiveness of skin assessments to identify these ulcers at POA. Beyond skin assessments, clinicians are using PrU prediction tools. The most common tool remains the Braden Scale for the Prediction of Pressure Sores.16 In recent years, 2 promising new prediction tools (Fragmment Score and Schoonhoven Prediction Rule) have been developed. However, those tools lack sufficient studies to evaluate their predictive validity.17,18
A recent meta-analysis of PrU risk assessment scales found the Braden Scale to be the best validated and a better risk predictor than nurses' best judgment, but the sensitivity was low and specificity adequate (57% and 67.5%, respectively) for a screening technique.19 Despite the widespread use of these techniques, the incidence of nosocomial pressure ulceration remains unchanged over the last 5 years.19 New strategies and innovative methods must be developed to drive down the incidence of pressure ulceration for the benefit of patients, and the Medicare rules provide strong economic incentives to do so.
The authors report the results of a pilot study using a TMI ImageMed System (Trillennium Medical Imaging, Inc [TMI], Holland, Ohio) to identify the at-risk patient and compared these data with Braden Scale scores as recorded simultaneously by unit nurses and specially trained wound care nursing personnel. The principal investigator (C.B.) and coinvestigator (D.J.) were funded by TMI to evaluate the TMI ImageMed System. To ensure the TMI ImageMed System was fully functioning, another coinvestigator (B.B.) was always on call for the research nurses. Finally, a PrU consultant (C.L.) and a biostatistician (K.F.) were employed to complete the research team. No data were reported to TMI until the study was completed at Duke University Medical Center, Durham, North Carolina, thus decreasing the potential for TMI to influence the results.
Design, Materials, Setting, and Participants
A prospective repeated-measures design was used to collect patient data. Research coordinators offered all patients admitted to the general medical service at Duke University Medical Center the opportunity to participate in this study until a population of 100 participants had completed the study. It took the research team 1.5 years to procure the 100 participants and complete the analysis. The only exclusionary criterion was the presence of a PrU on admission. Only adults were included because of higher incidence rates as compared with younger adults.
TMI ImageMed System
The 3 proprietary components of the system include the infrared camera, software, and the server/database.
The camera is the most recognizable component of the system and, in a sense, the most important. The camera takes 2 separate images. It captures 76,800 temperature readings, with a 0.06° C accuracy, and then generates a thermal image by assigning colored pixels to a specific temperature range. A visual light, or digital image, is captured by a visual light camera that is included in the device. Without the camera, TMI is unable to capture temperature data and image sets and generate the actual thermal image. TMI Systems provide the latest in FPA microbolometer technology and ensure the highest-quality, guaranteed class A detectors in imaging systems. TMI Systems demonstrate consistent and repeatable "black-body" accuracy: the key of infrared temperature evaluation.
TMI's software, TMI Image MHS 5.0, is a software package that allows scalability and flexibility in the clinical environment. TMI's software, in conjunction with the server/database, allows TMI to assess the suspected wound areas of a patient who is scanned by a TMI camera. The component that is least likely to be seen, but has the largest impact over time, is the server/database. The database is designed to work seamlessly with TMI's proprietary software to ensure that all Health Insurance Portability and Accountability Act-compliant patient data are accounted for, associated to the patient, the clinic, and the physician, and that all fields of the patient's medical history are searchable. The equipment itself is a Redundant Apple Dual 2.5-GHz OS X Server. The database software is MySQL, which is included and supported by the OS X Server. Consistency checks and removal of erroneous data are performed biweekly; backup to hot standby is automated daily. The data are transmitted via file transfer protocol to the RAID (computer servers). It can be interrupted and possibly copied along the path; however, the file itself is in a proprietary format, encrypted, and useless to anyone who intercepts the stream of data being recorded by the camera.
After obtaining informed consent, images of the sacral and heel surfaces were obtained using the TMI System Figure 1, which collects a phased array of tissue temperatures and a visual image of the site. The sacrum and heels were selected for continuous monitoring because a comprehensive review of the literature by the NPUAP found that these 2 anatomical locations were the most prevalent areas for PrU development in hospitals.20 Research nurses were trained on the use of the cameras and had to achieve at least a 95% interrater and intrarater reliability before they were allowed to collect scans. Both interrater and intrarater reliability testings were conducted every 3 months on the research nurses to decrease data collection drift. The authors wanted to be assured that the research nurses were consistently obtaining accurate Braden Scale scores and scans.
All participants were positioned in the lateral decubitus position for imaging and allowed to rest in this position for 3 minutes to compensate for potential reactive hyperemia, which may lead to false-positive results. All scans were obtained from sacral and heel areas in this position using a TMI ImageMed System. The research nurses were instructed to use the infrared camera cross hair to determine the center of the bony prominence for heels and sacrum. This step in the protocol ensured an increased level of accuracy during data collection. Only 1 reading of the skin temperature was obtained each day. A single image for each point of interest was taken during each imaging session (sacrum, left heel, right heel). It is important to note that the goal was to compare the temperature values of each image with the surrounding tissue during that particular examination. It is not necessary to note whether the temperature was consistently the same over the course of time or on subsequent days, as the algorithm is based on what is occurring with the patient in that moment in time. The sacrum and heels were not marked in any way. The camera display has a cross hair that denotes the center of the display. The cross hair was positioned over the bony prominence each time an image was captured to ensure that the nurses were imaging the same area each time.
Each 64-temperature matrix scan representing a 3 × 3-inch area overlying the bony prominences was selected for analyses. Although multiple ranges (3 × 3 inches, 5 × 5 inches, etc) can be used to scan, it was determined by the researchers that a 3 × 3-inch area represented the majority of participants' bony surface areas. Participants with patterns of injury (defined as a difference of 1.5° C between adjacent skin areas) were assigned to the scan high-risk group, whereas those without temperature differences were assigned to the scan low-risk group. A "pattern of injury" was determined to be any temperature reading that had a 1.5° C variance within the 3 × 3-inch target area. This temperature variance was validated in previous work of Sprigle et al.21 All data were uploaded electronically to a central server for analysis. Comparisons were made to evaluate risk assignments between these 2 groups. Both groups of nurses were blinded to the high- and low-risk scan groups.
A thorough Braden Scale for Predicting Pressure Sore (Braden Scale) risk assessment was performed by a team of research nurses who had received extensive training in completing the risk assessment tool as suggested by Braden et al.22 The 2 research nurses had greater than 5 years of clinical experience. A senior certified wound, ostomy, and continence nurse (WOCN) with more than 20 years of experience was employed to educate the research nurses on scoring the Braden Scale. She reviewed the Braden Scale with the research nurses. The research nurses were also required to watch a video developed by Braden et al23 on scoring patients using the Braden Scale. Moreover, the WOCN identified 10 randomly selected patients and scored each. She served as the criterion standard, given her extensive experience in PrU care. She then compared the scores of the research nurses. The research nurses had to achieve a 95% interrater and intrarater reliability before they were allowed to complete the Braden Scale on study participants. This procedure was repeated every 3 months between the WOCN and the 2 research nurses.
The Braden Scale score for each participant was also recorded by unit nurses on admission as per the existing unit policy and collected by the study nurses after completing their independent assessment. The Braden Scale scores were also calculated on each subsequent day by both groups of nurses (study nurses and unit nurses) until the participant was discharged. Participants with Braden Scale scores of 16 or less were assigned to the Braden Scale high-risk group, and participants with Braden Scale scores of greater than 16 were assigned to the Braden Scale low-risk group (consistent with unit policy). Imaging data were also collected on each subsequent day for each participant until the time of discharge from hospital. Positive images were recorded and monitored. Figure 2 Patient is a 91-year-old woman who showed no visual signs of erythema or tissuedamage on the sacral area. The image clearly shows patterns of injury associated with PrU development. The black spots outlined with blue signify a temperature variation of ±1.5° C. This has been the marker of indication that has shown up repeatedly. The graph listedsimply plots the temperature readings from the image and plots them along x and yaxes. Any indication of ±1.5° C will trigger an abnormal reading based on software parameters. illustrates one example of how an early scan-positive PrU appears with imaging independent of visual identification by the unit nurses.
Data were analyzed using SAS version 9.1 (SAS Institute Inc, Cary, North Carolina). Descriptive statistics were used to describe the participants. The authors also analyzed frequencies for categorical variables and means and SDs for Braden Scale scores between unit nurses and the research nurses.
For each assessment of a participant, an 8 × 8 matrix of temperature readings was produced. These data were brought into SAS. The minimum, maximum, 75th quartile, median, and mean were calculated for each sample. The authors defined that a temperature differential of 1.5° C in the matrix of greater than the surrounding tissue was indicative of increased risk of a PrU based on the research of Sprigle et al.21
To determine a 1.5° C differential, the authors compared 3 different calculations. The first calculation was to make the maximum temperature reading minus the minimum temperature reading in the 8 × 8 matrix and dichotomize it at the 1.5° C cutoff to low risk (0) and high risk (1). The second calculation was to subtract the minimum temperature reading from the 75th quartile and dichotomize to low- and high-risk using the 1.5° C cutoff, and the third method was to take the mean temperature minus the minimum temperature and dichotomize as described above. This dichotomous coding corresponds to the low risk (>16) and high risk (<16) used in the Braden Scale.
Analysis of Braden Scale Score
To assess differences in the Braden Scale administered by unit nurses and research nurses, the authors used mixed modeling to model differences between Braden Scale raters over time, using the Braden Scale as a continuous variable. The authors also dichotomized the Braden Scale by low risk (>16) and high risk (≤16) and used generalized estimating equations to look at differences in Braden Scale scores between unit and research nurses.
Analysis of Braden Scale Scores and Thermographic Scores
To examine the differences over time among the scores of the Braden Scale as compared with the infrared imaging scores, the authors dichotomized both the Braden Scale variable and the infrared imaging temperature change into low- and high-risk groups as previously described. The authors modeled the data using both generalized estimating equations and mixed modeling for nonnormal data. Compound symmetry was used as the covariance structure for the categorical mixed models.
These data are dependent on several levels. There are multiple measurements on participants over time, multiple measurements on participants by body site, and multiple readings on participants (nurse, research nurses, and thermographic data). To account for this dependence, it requires models that can account for the intraparticipant correlation. Two different modeling techniques were used to assess group and time differences. General estimating equations use robust SEs and an independent covariance matrix to account for dependence (repeated measurements on each person). The second method was using a mixed model for nonnormal data that can handle repeated measures of a dichotomous outcome in a random- and fixed-effects context. By using a random intercept, the within-participant correlation can be accounted for to obtain more appropriate parameter estimates. Generalized estimating equations and mixed modeling for nonnormal data provided similar results, leading to identical conclusions. For efficiency, the authors reported only statistics from the mixed models.
The linear mixed model for this can be written as follows:
accounts for the measurement error;
accounts for the dependence over time.
Calculating Odds Ratios
The model is ulcer = rater time rater × time, where ulcer is defined as a 0, 1 of low or high risk of an ulcer, rater is the type of rater (nurse, research nurses, and image set), and time is time. There are several measurements per person (multiple observations and multiple time points). The results are a summary of the odds ratio across all time points. The odds (estimation of probability) of infrared imaging are 6.8 times more likely to come up with a "high-risk" interpretation of getting a PrU compared with the unit nurse.
A total of 399 consecutively admitted participants were screened to obtain the 100 participants reported in this study. Of the 299 participants who refused to participate, there were no outstanding characteristics compared with those who participated. The major reason for refusing to participate in the study was the perception of patient and/or family members believing that the potential participant was "too sick" to be in a study (92%). Descriptive statistics were used to identify reasons for participants withdrawing during the study. The major reasons for participants withdrawing from the study were verbal request to withdraw from the study (n=5), refused to have images (n = 4), worsening medical conditions (n = 3), transferred to other units (n = 2), and severe pain on movement (n = 2). The mean age of the participants was 54.5 years (range, 23-92 years). There were 52 men and 48 women among the participants, with a mean length of stay on the medical unit of 3.93 days (range, 1-18 days). Only 5 of 100 participants developed PrUs (2 being at Stage II and 3 being at Stage I).
Imaging, using the 3 different algorithms, identified a substantial portion (22%-39%) of the authors' participant population as being at high risk for a PrU. Table 1 presents the number of observations categorized as being at high risk for a PrU based on a 1.5° C differential, based on 3 different methods for calculating the temperature differential. The first method is using the difference between the maximum and the minimum temperature of the 8 × 8 matrix. The second method uses the 75th percentile temperature value minus the minimum temperature, and the third method uses the mean temperature minus the minimum temperature. Using the first method, 39% of observations would be considered at risk for developing a PrU, whereas 28% and 22% would be identified as at risk using the latter 2 methods, respectively.
Table 2 presents the mean and median Braden scores for unit and research nurses, based on first observations, follow-up observations, and all observations combined. The mean score is lower among research nurses compared with unit nurses. To determine if these differences were significant, the authors used a mixed model to examine a nurse effect and a time effect and to determine if scores differ over time differently depending on type of nurse.
Based on results from the mixed model, there was a significant group effect (F = 12.21, P = .0006). There was neither time effect (F = .99, P = .54) nor rater effect by time effect (F = 1.10, P = .23). Thus, unit nurse raters consistently assigned higher Braden Scale scores as compared with research nurses. There was no difference over time on Braden scores, and Braden scores did not differ over time differently by rater. The authors categorized the Braden Scale scores into high and low risk as well. Interrater reliability between unit nurses and research nurses was poor. In examining data from the first observations only, there was 42% agreement (κ = 0.42). If all observations are used, the agreement decreases (κ = 0.40); however, this value does not account for multiple measures on the same subject (intraparticipant correlation). Comparing unit nurses, research nurses, and image set data, the image set data were more likely to classify an observation as high risk for a PrU compared with either the unit nurse or research nurse assessment using the Braden Scale score.
Table 3 presents the odds of the image sets (using upper quartile and mean minus max algorithms) classifying an observation as high risk compared with nurses using the Braden Scale. All participants noted to be at low risk by imaging were also noted to be at low risk by both groups of nurses.
PrUs remain an enormous cause of morbidity among patients admitted to acute care hospitals. Furthermore, the cost of providing care to patients who develop this complication after admission is considerable, and hospitals are no longer able to recover their costs of providing this care under new rules established in the most recent version of the Prospective Payment System for hospital services. Therefore, strong incentives exist to develop better strategies of prevention. Despite widespread use of risk assessment tools, such as the Braden Scale, little progress has been observed in reducing the incidence of pressure ulceration over the last decade.19 The present data shed some light on the wide variance on the prediction of PrUs using nurse experience versus the wide use of technology. There was considerable disagreement among scores obtained by the 2 groups of scorers: unit nurses routinely collecting Braden data in the busy world of an acute care hospital, versus a team with special expertise and dedicated time to collect information. The study's research nurses consistently rated patients at higher risk (lower Braden Scale score) compared with unit nurses. One explanation for the variance between the research nurses and the unit nurses may be the extensive education and reoccurring reliability testing (every 3 months) of the research nurses. Thus, the unit nurses may need more frequent education on completing the Braden Scale, especially when the health status of the patient changes.
Linking the imaging device and software to the hospital electronic medical record (EMR) affords the opportunity for ongoing surveillance of an entire inpatient census. Linked to the EMR, the imaging device can act as a surveillance tool, alerting clinical staff when a patient has not been scanned, or when thermal imaging suggests a potential pattern of injury, allowing perhaps the dispatch of an expert team to evaluate the patient and/or trigger prevention and treatment protocols. Time-stamped infrared images stored in a central server will likely lead to better documentation. In this study, there was lack of documentation of Braden Scale scores in several participants' records (data not shown). Thus, their data were eliminated from the analysis.
The ideal screening test would be both 100% sensitive and specific. The study device clearly identifies many more at-risk patients in this study than did the Braden Scale (higher sensitivity), although the numbers of participants who developed PrUs are not high enough to determine specificity. It is clear that larger studies must be done to further establish the operational characteristics of this device-notably, the positive and negative predictive values. The high rates of at-risk patients and anatomical locations identified by infrared imaging in this study provide an opportunity to provide location-targeted therapies, which might well provide significant savings.
The data suggest that the TMI ImageMed System was able to identify ulcers not identified by the nurses using the Braden Scale. Moreover, there were no additional ulcers that developed that were not detected by either the TMI system or Braden Scale. The imaging data accurately identified the anatomical location that was going to break down. Although this was a small study, the results are promising. It is not surprising that imaging identified a higher percentage of at-risk patients since it was identifying specific anatomical locations that were at risk. The device is extremely portable and can be used easily in the clinical setting. In addition, the hand held device can be incorporated into the screening process. Although scoring systems such as the Braden Scale predict that the entire patient might be at levels of risk, infrared imaging places risk at a specific anatomical location. Knowing the specific location(s) at risk allows the clinical staff to better target interventions. It should be noted that risk assessment scales should never replace clinical judgment.
With the graying of the US population, PrUs will only increase unless there is a shift to the use of objective and quantifiable technology to aid clinicians in identifying them earlier and to implement targeted preventive interventions. Although the cost of this technology has not been established, given the speed of scanning a patient, little in the way of increased personnel costs is anticipated. The present study suggests that the use of optical scanning technology may become the criterion standard for identifying PrUs earlier and hopefully preventing incipient PrUs in hospitals.
Further efforts to develop strategies to drive down the incidence in acute care hospitals and long-term-care settings are greatly needed. Technology will likely play an important role in this process, and the current study suggests an important role for advanced tissue imaging at the bedside.