Decreasing length and variation in the hospitalization process.8 Process control chart (see the "Methods" section) of 113 consecutive discharged patients' length of stay (LOS) (A) and its variation (B).8 Dashed lines represent upper confidence limits (UCLs) and lower confidence limits (LCLs), the boundaries of normal variation. Dotted lines separate probability zones (full standard deviations) labeled C, B, and A, moving away from the mean. CCP indicates clinical care pathway.
Morbidity and mortality outcomes. Comparison of 95% confidence intervals (error bars) for flap loss (A) and mortality (B) for this series and the flagship centers. HMO indicates health maintenance organization.
Hospitalization outcome. Comparison of 95% confidence intervals (error bars) for length of stay for this series' cohorts and the flagship centers.
Lueg EA. Comparing Microvascular Outcomes at a Large Integrated Health Maintenance Organization With Flagship Centers in the United States. Arch Otolaryngol Head Neck Surg. 2004;130(6):779-785. doi:10.1001/archotol.130.6.779
Copyright 2004 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2004
To determine if patients undergoing microvascular reconstructive head and neck surgery (MRHNS) at a large integrated health maintenance organization can expect outcomes similar to some of the best or flagship centers in the United States.
Outcomes (flap loss, mortality, length of stay), eligibility (recent consecutive US center experience), high-experience (100 cases), high-volume (26 cases per year), and flagship criteria were prospectively defined. A systematic MEDLINE search identified 17 eligible reports. Independent, blinded medical reviewers identified 5 centers (29%) as flagship centers.
The first 116 consecutive patients (average, 39 cases per year) who underwent MRHNS on this service.
All 5 flagship centers are major academic health centers ranked in the top 18 "best head and neck hospitals" in the United States. Flap loss (1.7% vs 4.4% for flagship centers; range, 0.9%-8.8%) and mortality (2.6% vs 2.8% for flagship centers; range, 0.5%-6.3%) rates were not significantly different. Although lengths of stay in flagship centers were similar to each other and the literature (mean, 21.4 days; range, 20.1-22.5 days), our length of stay was significantly shorter (8.8 days, P<.001).
For high-experience and high-volume centers, patients undergoing MRHNS at a large integrated health maintenance organization can expect morbidity and mortality outcomes similar to flagship centers in the United States, with shorter hospitalizations.
The quality of the tertiary surgery provided by major academic health centers (MAHCs)1 in the United States is recognized worldwide. On the other hand, Kaiser-Permanente, the nation's largest integrated (physicians and hospitals share financial risk2) health maintenance organization (HMO), now provides most tertiary surgery for its own members. In 1998, a dedicated microvascular reconstructive head and neck surgery (MRHNS) service was established to provide this for our members living in the Los Angeles (Calif) referral basin (south of Fresno and north of San Diego) who underwent oncologic resection at our Regional Head, Neck, and Skullbase Surgical Oncology Center. This service was the first high-volume (see flagship criteria) MRHNS service to be established outside an MAHC in the United States (Table 1).3- 7 The objective of this study was to determine if patients undergoing MRHNS at a large integrated HMO can expect outcomes similar to some of the best or flagship centers in the United States.
Roughly 1 in every 4 insured Southern Californians are members of Kaiser-Permanente, a large, nonprofit, group-model (patients choose their physicians from our group), integrated HMO, which provides comprehensive health care.2 The Southern California Permanente Medical Group is composed of more than 4000 multispecialty (partner and affiliated) physicians who are exclusively contracted to provide their health care at roughly 200 outpatient facilities and 12 major medical centers across Southern California. I performed all 117 free flaps (Table 2), most on our dedicated Flap Tuesday, with the other head and neck surgical oncologist (HNSO) (Michael P. McNicoll, MD) assisting. All patients were evaluated prospectively at our multidisciplinary (HNSOs and radiation oncologists) tumor board and received extensive preoperative medical and psychological preparation. As team leader, the patient's admitting HNSO managed the surgical issues. Medical issues were managed by an intensivist in the dedicated intensive care unit and a hospitalist on the dedicated ward. Discharge planners coordinated outpatient resources, including routine daily home nursing visits. After the first 57 patients, all the following patients were placed on a target 6-day inpatient clinical care pathway (CCP) (Figure 1).8 The study period began with the first service case on October 27, 1998, and was closed on the anniversary after having met the anticipated high-experience criteria. Therefore, the first 116 consecutive patients (average, 39 cases per year) who underwent MRHNS on this service between October 27, 1998, and October 27, 2001 were the subjects of this comparison (Table 3).
Flap loss (patients with complete flap loss) was selected because it is the most commonly used specific measure of reconstructive morbidity following MRHNS3- 7 and most microvascular surgeons believe it is an indicator (albeit relatively crude) of quality.9 Perioperative mortality (any cause within 30 days) was selected because it is the most commonly used measure of mortality following MRHNS3- 7 and most health care experts believe it is an indicator of quality.10 Length of stay (LOS) (complete contiguous days) was selected because it is the most commonly used measure of hospitalization following MRHNS5,11- 15 and most health care experts believe it is an indicator of resource cost.16
Similar (single US center, complete, consecutive, MRHNS experience reports of cases not restricted to a particular defect or donor site) and recent (published within 10 years of the close of this study) reports met the eligibility criteria. Flagship centers were defined as those centers that have published an eligible report that also satisfies both the high-experience (at least 100 consecutive cases) and high-volume (averages at least 1 case every few weeks or 26 cases per year) criteria. High-experience centers were selected, since most microvascular surgeons believe that greater experience is associated with improved outcomes.3- 7,9,11- 15 High-volume centers were selected, since a "broad body of evidence from large, population-based studies" has consistently demonstrated better outcomes for both major oncologic resections and high-risk vascular procedures.10 Once the outcomes, eligibility, high-experience, high-volume, and flagship criteria were specifically defined, a systematic MEDLINE search was undertaken. A review of the 722 citations containing the words free flap and head within their text identified 17 eligible reports. Two independent (no interest in the findings of this study) and blinded (no knowledge of the study's objective) medical reviewers evaluated the abstracts (text only with center identifiers removed) for the flagship criteria. Both independently excluded the same 12 reports. Therefore, 5 centers (29%) were defined flagship (Table 1).
We used a process control chart (PCC)8 to graph our LOS to analyze hospitalization (Figure 1). A PCC is simply a standardized statistical graph of a key quality characteristic (LOS in our case) of a given process over time. They have been used by the manufacturing sector for a long time to help decide whether a process is stable (only common, random, or normal variation is present) or unstable (a special cause of variation is also present). This decision is important because common-cause variation produces a normal (standard bell-shaped) distribution and is therefore predictable (within statistical limits). On the other hand, the decision that a process is unstable often leads to an investigation into the source of the special-cause variation. More recently, the usefulness of PCCs in helping to achieve continuous quality improvement in health care has been recognized.8 Upper and lower confidence limits are the absolute boundaries outside which a particular occurrence will be arbitrarily considered indicative of special-cause variation. These boundaries are conventionally set at 3 SDs from the mean to minimize the combined risk of mistakenly considering a truly common-cause occurrence to be indicative of special-cause variation (type I error, which may lead to tampering with a stable process) and visa versa (type II error, which may lead to undercontrolling an unstable process). It is extremely unlikely that a common-cause occurrence will lie outside these boundaries (type I error, 0.27%). The areas between these boundaries are also conventionally divided into 3 equal probability zones (labeled C, B, and A, moving away from the mean), which coincide with whole standard deviations. These zones are useful in identifying special-cause variation even if the occurrences lie within the upper and lower confidence limits. For example, 15 consecutive occurrences within 1 SD of the mean (zone C) is also extremely unlikely to occur if the process is stable.
The 95% confidence intervals (CIs) for flap loss and mortality were determined based on experience that they more closely approximate a binomial distribution.17 Significance was tested with a Poisson heterogeneity test (a goodness-of-fit χ2 test based on expected values from a weighted average Poisson model).18 The 95% CIs for LOS were determined based on experience that it more closely approximates a normal distribution.11 In addition to our overall LOS, the CIs for our sickest (American Society of Anesthesiologists [ASA] class >2) and irradiated subgroups were also determined to address concerns that these potential confounders might explain the significant differences in LOS. Since standard deviations for the flagship samples were not reported, we estimated them by multiplying the standard deviation of the largest (145 discharges) cohort from a flagship center reported19 by the ratio of the flagship's average over this cohort's average. Based on past experience with other LOS distributions, this scaling method usually provides a reasonable, yet conservative, estimate of standard deviation (Raoul Burchette, MS, written communication, October 31, 2002). Significance was tested with an independent-groups, unequal variance t test. Outcomes were extracted from the defining flagship report.3- 7 In several cases, LOS was not reported. In these cases, it was extracted from the largest flagship cohort reported.12- 14 Since these samples were smaller, their CIs were wider. All surgery and patient variables that were reliably extractable (either specifically stated or unambiguously derivable) from at least 2 defining flagship reports were compared directly with this series (Table 2 and Table 3). The 95% CIs for proportional differences (all except mean age) for this series and the flagship mean were calculated by the Newcombe-Wilson hybrid score method.20 Differences whose 95% CI did not encompass a zero difference were considered statistically significant. Significance for the difference in mean age was tested using a random-effects, 1-way analysis of variance test using standard deviations conservatively estimated from the sample ranges.21
All 5 flagship centers are MAHCs.1 In 2003, US News & World Report placed all these flagship centers among America's 18 "best hospitals for head and neck disorders."22 Four are also ranked among the 12 "best cancer hospitals," including the 2 highest ranked National Cancer Institute (NCI)–designated comprehensive cancer centers in the nation.22,23 Flap loss (1.7% vs a flagship average of 4.4%; range, 0.9%-8.8%; P = .24) and perioperative mortality (2.6% vs a flagship of average 2.8%; range, 0.5%-6.3%; P = .43) rates were not significantly different. These findings were confirmed by the overlap between the 95% CIs for our flap loss and mortality with the flagship centers (Figure 2). Although flagship LOSs were similar to each other and the literature (average, 21.4 days; range, 20.1-22.5 days),11 our overall (8.8 days, P <.001), sickest (9.2 days, P <.001), irradiated (9.6 days, P <.001), and pathway (7.1 days, P <.001) LOSs were all significantly shorter. These findings were confirmed by the wide separation between their and the flagship centers' 95% CIs (Figure 3). Our PCC (Figure 1) demonstrates that our hospitalization process is unstable (special-cause variation is present). All 56 consecutive LOSs after implementation of our CCP (57th discharge) is within 1 SD of the mean (zone C). Remember, even 15 consecutive occurrences within 1 SD are extremely unlikely to occur if the process is stable (see the "Methods" section).8 The variation graph confirms the significant variation dampening that occurred after implementation of the CCP (the main source of the special-cause variation).
Better surgical techniques generally improve outcomes over time. By the very nature of this study, our outcomes are more recent (Table 1). To minimize this inherent limitation, only recent (published within 10 years) reports were eligible. Although a 5-year cutoff would have diminished this effect even further, only 2 flagship centers (40%) would have remained. The M. D. Anderson Cancer Center, the top-ranked NCI-designated comprehensive cancer center in the nation, would have been 1 of those 3 centers excluded. However, the major improvements (microscope, instruments, flaps, training) in MRHNS techniques were already available before most (more than 90%) flagship patients underwent their surgery.3- 7 Furthermore, greater center and surgeon experience also improves MRHNS outcomes over time.3- 7 Certainly, our own outcomes, as well as the flagship centers', are better today. However, most of the flagship series include a significantly smaller percentage of early (first 100) flaps (85% vs 43% for the flagship centers, P<.05), which might have skewed our outcomes adversely.3- 7
Still, flagship patients underwent surgery on average 6.3 years earlier, and there has been a recent trend nationally (especially in California) to decrease LOS.16 One flagship center has already reported a significant reduction for a later cohort.13,19 It is very likely that the average LOS at all the flagship centers is significantly shorter today. However, our own LOS since implementation of a target 6-day inpatient CCP is now less than 1 week with concurrent improvements in a wide array of safety (morbidity and mortality) outcomes (Figure 3).24 A systematic re-review of the original 722 MEDLINE citations containing the text words free flap and head found the shortest average LOS for a large (at least 100 cases) cohort of MRHNS patients to be 11.2 days.11 Our CCP patients' (now more than 150 patients) LOS is significantly shorter than this (6.3 vs 11.2 days, P<.001).24
Of the demographic variables extractable from at least 2 defining flagship reports (Table 3),3- 7 mean age was significantly different. Indicators of socioeconomic status (income, education, employment, neighborhood, insured, minorities), which are often adversely associated with health status (comorbidities) and hence postoperative morbidity and mortality, were not reported. A single flagship center reported its percentage of nonwhites, which was significantly higher for this series (62% vs 18% for the flagship center, P <.05).4 Our patients were also significantly older (5 years) on average (60.2 vs 55.2 years for the flagship centers, P <. 001). This difference reflects our conviction that one's chronologic status (age) on its own is not an independent (directly) associated risk factor. This is underscored by the fact that we report the first MRHNS case in a nonagenarian patient (vigorous 95-year-old discharged on day 6).3- 7 However, advanced chronologic status is consistently associated with poorer health and hence has been dependently (indirectly) associated with greater postoperative morbidity,3,5- 7,15 mortality,3,5,6 and LOS15,24 in MRHNS patients.
Most experienced microvascular surgeons would agree that a patient's health status is the most powerful independent predictor of morbidity, mortality, and LOS following MRHNS.3- 7,11,15,25 Unfortunately, only 2 flagship centers reported any direct measure of it, the relatively crude and nonspecific ASA class (Table 3). Our percentage of sicker patients (ASA class >2) fell between the 2 reported (53% vs 32% and 74% for the flagship centers, P >.05).4,6 However, even if we completely exclude our healthier patients, the 95% CI for the LOS of our sickest (ASA class >2) patients, all with severe or life-threatening systemic disease, would still remain widely separated from the flagship CIs (Figure 3). Yet this does not entirely eliminate the possibility that a more sensitive and specific measure of health status for MRHNS patients, such as the Washington University Head and Neck Comorbidity Index (our mean was 0.54),26 might reveal that the flagship patients, although chronologically younger, were still sicker.
Of the disease (morbidity) or tumor status variables extractable from at least 2 defining flagship reports, malignant tumors (percentage of squamous cell carcinomas [SCCs]), size (percentage with American Joint Committee on Cancer [AJCC] primary tumor stage 3 or 4), and irradiated were significantly different (Table 3). Unfortunately, the upper aerodigestive tract sites were partitioned inconsistently (ie, oropharynx was grouped with the oral cavity in some flagship reports and pharynx in others), limiting reliable extraction of comparable mucosal subsite data. The malignant tumors in our patients were significantly more likely to be SCCs (95% vs 81% for the flagship centers, P <.05). The difference seems to reflect a higher-than-expected contribution of unusual tumor findings (sarcomas, basal cell carcinomas, non-SCCs) at 2 flagship centers (23% and 23%). The AJCC classifiable SCC primary tumors in our patients were also significantly more likely to be large (97% vs 79% in the flagship centers, P <.05). This difference reflects the fact that we are the designated regional tertiary-referral surgical oncology center with a dedicated MRHNS surgeon. One flagship report3 addressed pathology and site and found no association with flap loss. To my knowledge, no report has addressed size (or stage), since most MRHNS patients have large (advanced) primary tumors. However, I know of no microvascular surgeon who believes that patients with large SCC tumors might have more favorable outcomes compared with patients with smaller non-SCC tumors.
Significantly fewer of our patients underwent prior irradiation (12% vs 36% for the flagship centers, P <.05). This reflects the fact that our prospective multidisciplinary tumor board generally recommends primary oncologic resection followed by adjuvant radiation for advanced (AJCC clinical stage III or IV) resectable tumors. Therefore, we perform MRHNS on relatively few patients in whom irradiation fails, since in most cases aggressive oncologic resection has also failed. However, the 95% CI for the LOS of our irradiated patients remains widely separated from the flagship centers' CIs (Figure 3). Furthermore, there are now at least 9 large, peer-reviewed MEDLINE reports, 7 from flagship centers, involving almost 5000 free flaps that have specifically addressed the prior irradiation issue, and none have found that it was a risk factor for any of our outcomes.3- 5,7,11,27- 30
This series reflects our basic techniques (Table 2). Like most of the flagship centers we avoid (if possible) secondary (delayed) reconstructions, nonworkhorse (infrequently selected) donor sites, use of vein grafts, and planned simultaneous (multiple) donor site reconstructions. Excluding one statistical outlier (>3 SDs from the mean), our reported composite (bone flap) reconstruction rate was similar to the other flagship centers (26% vs 27% for the flagship centers; range, 23%-30%; P >.05). Our workhorse armamentarium (frequently [at least 10%] selected donor sites) includes the radial forearm, fibular, and anterolateral thigh. All 3 have long, large vascular pedicles that have helped us avoid the need for vein grafts entirely. Use of the infrequently selected but multipaddle subscapular composite torso donor site3 has also helped us entirely avoid the need for planned multidonor site reconstructions and the significant additional functional and aesthetic morbidity they inflict.
Where we do differ significantly with the flagship centers is our avoidance of torso donor sites (3% vs 52% in the flagship centers, P <.05) in lieu of extremity donor sites (97% vs 48% in the flagship centers, P <.05) if possible (Table 2). This difference reflects my experience that torso sites tend to be closer to the ablative site, require some repositioning or are less readily exposed, are inconsistently bulky, have shorter and deeper pedicles, and inflict greater donor morbidity than extremity sites like the radial forearm and fibula. Our embrace of the anterolateral thigh donor site for extensive soft tissue defects (19% vs 0% in the flagship centers, P <.05), in lieu of the more conventional rectus abdominis and latissimus dorsi donor sites (1% vs 21% in the flagship centers, P <.05), is emblematic of this approach. This extremely versatile, extensive soft tissue donor site is farther from the ablative site, requires no repositioning, is readily exposed, is moderately and uniformly thick, has a long superficial pedicle, is simple to harvest (with a cuff of medial vastus lateralis), and inflicts minimal donor morbidity.31 Although our preferred workhorse donor sites may be significantly different, we concur with an evolving consensus at many of the flagship centers that when it comes to the total number in one's armamentarium less may be more.4- 7
The objective of this study was to compare our outcomes with some of the best available. The implicit assumption was that high-experience and high-volume centers would have some of the best outcomes.3- 7 The fact that all 5 flagship centers turned out to be highly ranked MAHCs1,22,23 suggests that the selection methods were sound. Most knowledgeable microvascular surgeons would agree that (alphabetically) the M. D. Anderson Cancer Center in Houston, Tex, Memorial Sloan-Kettering Cancer Center and Mount Sinai Medical Center in New York, NY, University of California Medical Center in Los Angeles, and University of Pittsburgh Medical Center, Pittsburgh, Pa, have some of the best MRHNS outcomes in the United States. However, most would also agree that integrated HMO and MAHC patient populations are different.1,2 Some of these differences may be potential confounders that skewed our outcomes favorably or adversely (see the "Results" section). Yet our outcomes were only selected indicators of quality and cost.8- 11,16 For example, although mortality may be similar for 2 groups of patients, most surgeons would agree that quality was better if one group was sicker before surgery.8,10,18 This study does not permit fair conclusions (and none are made) regarding quality or cost, since several likely confounders could not be adequately controlled (statistically adjusted)18 with the flagship information available.
On the other hand, the size (243% longer), strength (widely separated CIs), and stability (separation remains even for sickest and irradiated strata) of the difference in LOS all suggest that it is robust. However, analysis of the flagship LOS based on extrapolation of our subgroup variation for 3 important risk factors (poorer health status, large tumor status, and prior radiation status) suggest that measured differences in case mix do not appear to adequately explain the large observed differences in LOS (Figure 3). Therefore, it is most likely that this robust difference is the product of one or more very powerful (many-fold stronger than health status) unmeasured confounder(s). An obvious candidate, by virtue of most MAHCs' public mission of "caring for those who remain uninsured,"1 might be a greater percentage of poor patients. However, like chronologic status, indicators of socioeconomic status have consistently been adversely associated with health status and hence in large part only dependently (indirectly) with poorer surgical outcomes. Independent risk factors are by definition more powerful than the risk factors dependent on them. For this reason, chronologic and socioeconomic status are generally less powerful than health status, the usual underlying independent risk factor. Another obvious candidate might be resident teaching status.1 However, we currently sponsor 16 accredited residency programs, including medicine, surgery, obstetrics/gynecology, pediatrics, and family practice (167 postgraduates). This is greater than the minimum to be considered an MAHC (4)1 and well within the flagship range (8-74).32
I believe, based on my experience (as a MRHNS fellow at a flagship center), that the most important unmeasured confounder is a relative disconnection of financial risk between admitting physicians and their hospitals. This disconnection is related to the fact that most HMOs (health insurance plans) today are both nonintegrated and for profit (plan accepts risks and rewards).2 Therefore, physicians and hospitals are insulated from each other's risk by the virtual cornucopia of intervening HMOs that insure their patients. The very nature of our nonprofit, group-model, integrated comprehensive health care delivery system2 means that all provider (physician and hospital) costs must ultimately come from the same pot (members' prepaid premiums). Therefore, we all have a direct stake in the financial fitness of our hospitals (and yes, even our health plan). Make no mistake. We are physicians first, and therefore achieving high quality for our patients is always the first and most important consideration. However, as vested participants, achieving high value (quality over cost) for our entire organization is also an important consideration. It is for this reason that every MRHNS patient was evaluated every workday by their admitting HNSO, who was personally motivated to achieve safe early discharge.11,16,24 It is this same motivation that has led to implementation of several well-established efficiency measures (see the "Methods" section) and perhaps most importantly a target 6-day inpatient CCP.24
I remind the reader that LOS was only a selected indicator of cost. Although we believe our approach increases value, since hospitalization remains the "big ticket" item in health care,16 this may not necessarily be true. Prospectively controlled, longitudinal studies that use more direct measures of global costs would be required (although probably impractical) to definitively address the issue of cost across the integrated HMO and MAHC settings. However, this study at least suggests one important benefit of integration (physicians and hospitals sharing financial risk). We are able to return our patients safely home to their families sooner.
In conclusion, for high-experience and high-volume centers, patients undergoing MRHNS at a large integrated HMO can expect morbidity and mortality outcomes similar to flagship centers in the United States, with shorter hospitalizations.
Corresponding author and reprints: Edgar A. Lueg, MD, FRCSC, 4900 Sunset Blvd, Suite 6C, Los Angeles, CA 90027 (e-mail: firstname.lastname@example.org).
Submitted for publication February 17, 2003; final revision received October 27, 2003; accepted November 25, 2003.
I thank Raoul Burchette, MS (senior data consultant, Research and Evaluation, Kaiser-Permanente), for his statistical advice and guidance throughout this project. I also thank Michael P. McNicoll, MD (senior HNSO), Frederic R. DiTirro, MD (chief of head and neck surgery), and Robert A. Kagan, MD (chief of radiation oncology), for their support.