Probability can be determined by drawing a vertical line connecting the value of each variable with the point score at the top of the diagram. The scores for each variable are then summed to give a total points score, which is plotted along the total points line at the bottom of the nomogram. This line is projected to the probability of in-hospital mortality of the patients with head and neck cancer who are hospitalized.
eTable. Multilevel analysis estimating in-hospital mortality for individual comorbidities that make up the comorbidity score, (n = 72,114)
Customize your JAMA Network experience by selecting one or more topics from the list below.
Identify all potential conflicts of interest that might be relevant to your comment.
Conflicts of interest comprise financial interests, activities, and relationships within the past 3 years including but not limited to employment, affiliation, grants or funding, consultancies, honoraria or payment, speaker's bureaus, stock ownership or options, expert testimony, royalties, donation of medical equipment, or patents planned, pending, or issued.
Err on the side of full disclosure.
If you have no conflicts of interest, check "No potential conflicts of interest" in the box below. The information will be posted with your response.
Not all submitted comments are published. Please see our commenting policy for details.
Adjei Boakye E, Osazuwa-Peters N, Chen B, et al. Multilevel Associations Between Patient- and Hospital-Level Factors and In-Hospital Mortality Among Hospitalized Patients With Head and Neck Cancer. JAMA Otolaryngol Head Neck Surg. Published online March 19, 2020. doi:10.1001/jamaoto.2020.0132
Are there patient- and hospital-level factors associated with in-hospital mortality among hospitalized patients with head and neck cancer?
In this cross-sectional study of a nationally representative population of 85 440 patients with head and neck cancer, prevalence of in-hospital mortality was 4.2%. Male sex, increased number of comorbidities, metastatic cancer, nonelective hospital admission, weekend hospital admission, and admission to a nonteaching hospital were characteristics associated with higher odds of in-hospital mortality.
Both patient- and hospital-level factors are associated with in-hospital mortality among hospitalized patients with head and neck cancer.
Risk factors for in-hospital mortality of patients with head and neck cancer (HNC) are multilevel. Studies have examined the effect of patient-level characteristics on in-hospital mortality; however, there is a paucity of data on multilevel correlates of in-hospital mortality.
To examine the multilevel associations of patient- and hospital-level factors with in-hospital mortality and develop a nomogram to predict the risk of in-hospital mortality among patients diagnosed with HNC.
Design, Setting, and Participants
This cross-sectional study used the 2008-2013 National Inpatient Sample database. Hospitalized patients 18 years and older diagnosed (both primary and secondary diagnosis) as having HNC using the International Classification of Diseases, Ninth Revision, Clinical Modification codes were included. Analysis began December 2018.
Main Outcomes and Measures
The primary outcome of interest was in-hospital mortality. A weighted multivariable hierarchical logistic regression model estimated patient- and hospital-level factors associated with in-hospital mortality. Moreover, a multivariable logistic regression analysis was used to build an in-hospital mortality prediction model, presented as a nomogram.
A total of 85 440 patients (mean [SD] age, 62.2 [13.5] years; 61 281 men [71.1%]) were identified, and 4.2% (n = 3610) died in the hospital. Patient-level risk factors associated with higher odds of in-hospital mortality included age (adjusted odds ratio [aOR], 1.03 per 1-year increase; 95% CI, 1.02-1.03), male sex (aOR, 1.23; 95% CI, 1.12-1.35), higher number of comorbidities (aOR, 1.14; 95% CI, 1.11-1.17), having a metastatic cancer (aOR, 1.49; 95% CI, 1.36- 1.64), having a nonelective admission (aOR, 3.26; 95% CI, 2.83-3.75), and being admitted to the hospital on a weekend (aOR, 1.30; 95% CI, 1.16-1.45). Of the hospital-level factors, admission to a nonteaching hospital (aOR, 1.48; 95% CI, 1.24-1.77) was associated with higher odds of in-hospital mortality. The nomogram showed fair in-hospital mortality discrimination (area under the curve of 72%).
Conclusions and Relevance
This cross-sectional study found that both patient- and hospital-level factors were associated with in-hospital mortality, and the nomogram estimated with fair accuracy the probability of in-hospital death among patients with HNC. These multilevel factors are critical indicators of survivorship and should thus be considered when planning programs or interventions aimed to improve survival among this unique population.
There were an estimated 14 620 deaths due to head and neck cancer (HNC) in 2019.1 Known clinical factors associated with HNC mortality include stage of presentation,2,3 anatomic subsite,2 and treatment modality.3,4 Treatment modality is often a function of clinical stage of disease at presentation; for most cases of HNC, treatment typically includes surgery. Owing to the complexity of the head and neck anatomy and tumor factors, surgical treatment procedures are extensive and are often associated with postsurgical complications that require hospitalization of patients for an extended time.5,6 Even nonsurgical treatments such as chemoradiation are often associated with short-term toxicities that lead to hospitalizations.7 Prolonged hospitalization of patients with HNC is associated with significant health care costs8 and may even have negative health outcomes.9 Thus, it is important to understand risk factors for in-hospitality mortality to optimize health care outcomes in this highly vulnerable population.
Factors that affect mortality of patients with HNC in the hospital may be patient-related, tumor-specific, hospital-related, or other. Previously reported patient characteristics associated with higher cancer mortality include advanced patient age,10 lower socioeconomic status,10 minority race,11 male sex,11 and Medicaid status or lack of insurance.10,12-15 Cancer-specific attributes, such as advanced stage of presentation, and nonsurgical treatments have been associated with increased risk of mortality in patients with HNC.16 Other studies have also reported hospital volume,17-20 hospital type characterized as academic or nonacademic,21 and institutional surgeon volume22 as significant predictive factors of mortality in patients with HNC.
However, there is a paucity of research exploring multilevel correlates of mortality in hospitalized patients with HNC. It is important to elicit which patient and hospital factors affect mortality so treatments are tailored based on the patient’s clinical profile and access to cancer care. Therefore, the objective of our study was to (1) examine multilevel factors associated with in-hospital mortality and (2) construct a nomogram to predict the risk of in-hospital mortality among hospitalized patients diagnosed with HNC.
This study was a cross-sectional analysis of hospital discharge data from the National (Nationwide) Inpatient Sample (NIS) database between January 2008 and December 2013. The NIS data were obtained from the Healthcare Cost and Utilization Project of the Agency for Healthcare Research and Quality.23 The NIS is the largest publicly available all-payer inpatient care database in the United States containing data on more than 7 million hospital stays each year (more than 30 million weighted hospitalizations nationally). Before 2012, the NIS represented a 20% stratified sample of all community nonfederal US hospitals. Starting in 2012, the NIS was redesigned to represent 20% of all discharges from nonfederal US hospitals, representing 94% of all discharges from US community hospitals. Detailed information on the NIS design and sampling methods can be obtained at https://www.hcup-us.ahrq.gov. The NIS is a publicly available database containing deidentified patient information and was deemed exempt from review by the institutional review board at Saint Louis University in St Louis, Missouri. Patient consent was also waived because data were publicly available.
Patients 18 years and older with a diagnosis (both primary and secondary) of malignant neoplasms of lip, salivary gland, gums, floor of mouth, other parts of the mouth, oropharynx, nasopharynx, hypopharynx, posterior tongue, anterior tongue, other parts of the tongue, other and ill-defined sites, and larynx were identified using International Classification of Diseases, Ninth Revision, Clinical Modification diagnostic codes for the study (Table 1).
The primary outcome of interest was in-hospital mortality, defined as mortality that occurred during hospitalization as coded from the discharge disposition of patient (alive or dead). Patient- and hospital-level characteristics that are associated with in-hospital mortality were selected based on previous literature.10,15,19 Patient-level characteristics examined included age (accessed as a continuous variable), sex, race/ethnicity as defined in NIS data (white, black, other, missing), insurance type (private, Medicare, Medicaid, other [included self-pay, no charge, workers’ compensation, the Civilian Health and Medical Program of the Uniformed Services, the Civilian Health and Medical Program of the Department of Veterans Administration, Title V, and other government programs]), residential region (Northeast, Midwest, South, West), admission type (elective/nonelective), admission day (weekend/weekday), metastasis (yes/no), indicators for whether chemotherapy or radiation (yes/no) were performed during the hospitalization, and anatomic sites. Postoperative complication was defined using International Classification of Diseases, Ninth Revision, Clinical Modification codes associated with mechanical infection, bleeding, and wound dehiscence. Residential income was estimated based on the median household income at the zip code level and was classified into quartiles with values ranging from 1 for low-income zip code (first quartile) to 4 for high-income zip code (fourth quartile). Length of stay (LOS) was calculated by subtracting the admission date from the discharge date, with same-day stays coded as 0. Comorbidities were classified using the Elixhauser Comorbidity Index24 and the final variable was a summation of number of comorbid conditions defined by the Elixhauser Comorbidity Index. Hospital-level characteristics examined included hospital teaching status (nonteaching/teaching) and hospital bed size (small, medium, large).
Weighted proportions for categorical variables and means (SDs) for continuous variables were used to describe the study population overall and stratified by anatomic sites. Then, weighted univariable and multivariable multilevel hierarchal regression analyses, using generalized linear mixed models, were used to examine the association between patient- and hospital-level characteristics and in-hospital mortality. We required hospitals to have at least 10 patients diagnosed as having HNC for inclusion in the analysis to appropriately measure hospital-level variability (resulting in a sample size of 72 114 for the multilevel hierarchal regression analyses). Analyses were performed using SAS, version 9.4 (SAS Institute) procedures, which incorporate survey sampling weights to account for the complex sampling design used in NIS and to provide representative estimates of the US population. All analyses were 2-tailed and statistical significance was determined by whether a CI excludes 1.
For the nomogram, only variables that were significantly associated with in-hospital mortality and clinically relevant were included in the model. The area under the receiver operating characteristics curve was calculated to determine the accuracy of the model’s predictive value. Receiver operating characteristics areas of 0.6 to 0.7, 0.7 to 0.8, 0.8 to 0.9, and 0.9 to 1 represent a poor, fair, good, and excellent discriminative model, respectively. The nomogram was developed using R, version 3.2.2 (R Foundation for Statistical Computing).
Table 2 summarizes the characteristics of 85 440 patients who were hospitalized between 2008 and 2013 for HNC as well as the hospital-level characteristics. The mean (SD) age of patients was 62 (14) years and the mean (SD) LOS in the hospital was 6.7 (8.1) days. Patients with HNC had a mean (SD) of 2 (1.7) comorbidities. Most patients were men (61 281 [71.7%]), white (55 630 [65.1%]), admitted on a weekday (72 575 [84.9%]), and had nonelective admission (51 459 [60.2%]). Among all HNCs, the anatomic sites that had the highest percentage of hospitalization were larynx (22 628 [26.5%]), followed by oropharynx (12 942 [15.1%]) and other parts of the tongue (7896 [9.2%]). Large-bed hospitals and teaching hospitals accounted for most hospitalizations. The prevalence of in-hospital mortality was 4.2% (n = 3610).
Patient- and hospital-level characteristics of all hospitalized HNCs stratified by anatomic sites are provided in Table 3. For all anatomic sites, the proportion of men was higher than women except for anterior tongue and gum where the proportion was more comparable between the sexes. The proportion of patients with nonelective admissions was higher than elective admissions for all anatomic sites except anterior tongue and salivary glands where elective admissions was higher and there was equal proportion for other parts of the mouth site. A higher proportion of patients were admitted to teaching hospitals for all anatomic sites except ill-defined sites where a higher proportion of patients were admitted to nonteaching hospitals.
The results of the univariable and multivariable multilevel hierarchal models for estimating in-hospital mortality are summarized in Table 4. In the adjusted model, the patients’ odds of in-hospital mortality increased if they were older (adjusted odds ratio [aOR], 1.03 per 1-year increase; 95% CI, 1.02-1.03), had longer stay at the hospital (aOR, 1.02 per 1-day increase; 95% CI, 1.01-1.03), and had a higher number of comorbidities (aOR, 1.14 per 1-unit increase in Elixhauser Comorbidity Index score; 95% CI, 1.11-1.17). Compared with women, men (aOR, 1.23; 95% CI, 1.12-1.35) had significantly higher odds of in-hospital mortality. Similarly, compared with patients admitted to the hospital on a weekday, those admitted to the hospital on a weekend (aOR, 1.30; 95% CI, 1.16-1.45) had a significantly higher likelihood of in-hospital mortality, as were patients who had metastatic cancer (aOR, 1.49; 95% CI, 1.36-1.64) or postoperative bleeding (aOR, 1.92; 95% CI, 1.36-2.70) compared with those without metastasis or postoperative bleeding, respectively. Patients with a nonelective admission (aOR, 3.26; 95% CI, 2.83-3.75) had significantly higher odds of in-hospital mortality compared with those who had elective admission. At the hospital level, admission to a nonteaching hospital (aOR, 1.48; 95% CI, 1.24-1.77) was associated with significantly higher odds of in-hospital mortality compared with admission to a teaching hospital.
Looking at individual comorbidities that make up the Elixhauser Comorbidity Index score, patients with congestive heart failure (aOR, 1.58; 95% CI, 1.35-1.86), pulmonary circulation disorder (aOR, 2.48; 95% CI, 1.97-3.14), paralysis (aOR, 1.81; 95% CI, 1.35-2.42), kidney failure (aOR, 1.26; 95% CI, 1.06-1.51), liver disease (aOR, 1.64; 95% CI, 1.35-1.99), coagulopathy (aOR, 2.21; 95% CI, 1.90-2.57), weight loss (aOR, 1.52; 95% CI, 1.37-1.70), or fluid and electrolyte disorders (aOR, 1.70; 95% CI, 1.55-1.87) had significantly higher odds of in-hospital mortality (eTable in the Supplement).
For the nomogram development, the following predictors were examined: age, sex, race/ethnicity, insurance type, admission type, metastasis, comorbidities, anatomic site, and hospital teaching status. The multivariable logistic regression model used to construct the nomogram had an area under the receiver operating characteristics curve of 0.72 and therefore classified as fair. The nomogram provides a practical tool for health care professionals to assess the risk of death of patients. As an example, using the points scale in the Figure, a black (9 points) man (8 points) aged 70 years (62 points) with Medicaid insurance (8 points) has 6 comorbidities (28 points), metastatic cancer (12 points), and nonelective admission (42 points) at a nonteaching hospital (12 points) and therefore scores a total of 181. The lower scale of the nomogram shows that this score corresponds to a risk of in-hospital mortality of about 0.15 (15%).
In this study, we aimed to assess multilevel factors associated with in-hospital mortality among hospitalized patients with HNC. We found that patient-specific parameters associated with in-hospital mortality included age, sex, LOS, type of admission, admission day, metastasized tumor, and number of comorbidities. We also found that the teaching status of an institution was the most significant hospital-level parameter associated with in-hospital mortality. In addition, we developed a nomogram that estimates the individual risk of in-hospital mortality based on sociodemographic, clinical, and hospital factors.
To our knowledge, this is the first risk-prediction nomogram attempting to evaluate in-hospital mortality in hospitalized patients with HNC. Age has the highest score in the nomogram followed by number of comorbidities, admission type, and insurance type. Nomograms are currently the most accurate available tools, with the greatest discriminating characteristics for predicting outcomes in patients with different cancers.25-27 They provide easily understood outcome probabilities and individualized disease-related risk estimations that facilitate patient management–related decisions. Nomogram use may not only facilitate early management decision-making but may also minimize unnecessary tests and expenses. The traditional odds ratios give an estimate of relative difference of odds on a multiplicative scale. Although this is informative in interpreting risk factors, it does not give a direct relationship between risk factors and the probability of an event (in-hospital mortality) because the outcome-predictor association is not linear. A nomogram is advantageous in that users can tell the association of a risk factor with the outcome by simply adding the points.
Consistent with previously published studies, we found that demographic factors such as patient age and sex are independent risk factors for in-hospital mortality in both the nomogram and adjusted multilevel analysis.28-32 Age had the highest score in the nomogram, indicating that older patients have a higher in-hospital mortality risk. The multilevel analysis also showed that each additional year a patient ages, their in-hospital mortality risk increases by 3%. Older patients may present with more advanced diseases that limit definitive treatments.33 In addition, increased age has a direct association with other health-related outcomes such as hospital complications, LOS, and intensive care unit admissions.34,35 This can often lead to undertreatment of cancer in older groups owing to fear of intolerance and adverse effects, resulting in worse cancer-specific outcomes. The age discrepancy in HNC in-hospital mortality observed in our study may also be explained by the increasing prevalence of human papillomavirus–positive HNC. Compared with their older counterparts, younger patients are more likely to have human papillomavirus–associated oropharyngeal cancer, which has a very different etiology and pathophysiology and carries a much better prognosis than those of non–human papillomavirus–associated HNCs.36-38 Our results show that female patients diagnosed as having HNC have a survival advantage compared with male individuals, a finding consistent with other studies.11,39 Male individuals exhibit more risk factors such as smoking and alcohol use, which are associated with worse treatment outcomes and higher comorbidities.11 Moreover, once diagnosed, male patients are less likely to seek medical care and are less compliant with treatment protocol compared with their female counterparts.11
We found that patients with higher number of comorbidities had higher odds of in-hospital mortality, with each additional comorbidity increasing in-hospital mortality rate by 14%. Specifically, patients with pulmonary circulation disorders, coagulopathy, paralysis, congestive heart failure, liver disease, or fluid and electrolyte disorders had the highest odds of in-hospital mortality. Similarly, in the nomogram, number of comorbidities had the second highest score. Patients with HNC have one of the highest comorbidities of all cancers, only second to colorectal and lung cancers.28 The presence of comorbidities has long been established as an independent prognostic factor in HNC because they can affect diagnosis, treatment modality and severity, and response to treatment.28 Comorbidities are associated with worse survival outcome in almost all HNC subsites, including oropharyngeal, oral cavity, early laryngeal, and nasopharyngeal cancers.40 Some studies have also found associations between higher comorbidities and cancer-specific survival in HNC owing to higher rates of disease recurrence28 and more aggressive tumor.41 The most commonly affected organ systems outside the tumor site of interest are cardiovascular and pulmonary.42
Our study included type of admission and day of admission to better determine optimal admission conditions. Examining differences between nonelective and elective admissions indicated that patients who were admitted under emergent conditions have higher risk of in-hospital mortality. Admission type was the third highest score in the nomogram, meaning it has a high effect on in-hospital mortality among hospitalized patients diagnosed with HNC. Surgical procedures performed during emergency admissions exacerbate risk of mortality.43 Tracheostomy and reconstructive surgeries have the highest mortality rates owing to increased postoperative conditions, such as respiratory complications and wound infections.43 In addition, our results showed an inverse association between admission during the weekend and in-hospital mortality. Reduced number of staff on the weekend may result in delayed diagnosis and management of a patient’s condition, placing further emphasis on the importance of robust multidisciplinary team management, including nursing staff, pharmacists, and physicians.18 Furthermore, patients admitted during the weekend are more likely to be nonelective, emergency admissions.44
Hospital-associated factors also play a significant role in in-hospital mortality of patients with HNC. Previous studies have shown that hospitals with lower HNC mortality rates tend to be academic centers19 with high surgical case volumes.17,18,22 Similarly, we found that patients with HNC treated at teaching or academic institutions have lower mortality risk compared with those treated at nonacademic hospitals in both the nomogram and multilevel analysis. Teaching hospitals are more likely to be high-volume centers and perform more surgeries, both ablative and reconstructive, compared with community or nonteaching institutions, which are more likely to treat HNC with radiation or without chemotherapy.45 Teaching/academic institutions have been reported to have better compliance with National Institutes of Health guidelines,17,46,47 which may be attributed to more frequent surveillance and familiarity with guidelines. Moreover, academic centers tend to have more abundant multidisciplinary teams, specialists, and ancillary staff that are better equipped to minimize and manage perioperative and postoperative complications.17 This may also explain improved survival of patients with cancer of the larynx and mucosal sites, which have complicated postoperative courses, at academic institutions.45,48
We found that patients with HNC with greater LOS at the hospital had greater risks of in-hospital mortality. Length of stay has been associated with posttreatment outcomes in other cancers. For instance, colorectal cancer surgery routinely uses LOS as a quality indicator, with longer LOS linked to higher postoperative complications, advanced age, and more comorbidities.49,50 Moreover, lengthier hospitalizations are associated with increased risk of hospital-associated complications such surgical site infections, ventilator-associated pneumonias, and clinician errors. Length of stay is an important component of health care cost and an indicator of health-related quality of life.51 We must also consider the role of population bias because patients who warrant more extended hospitalizations tend to be sicker in general.
The current study has limitations. We used administrative discharge data, and the data set did not have information regarding disease stage or tumor size. Thus, cancer severity was not considered in our study. It is well known that patients who have advanced-stage cancers have poorer outcomes. However, limiting the data to patients who received inpatient care may have limited some of the variation in disease severity because all of those patients had disease that was at least severe enough to warrant inpatient treatment. In addition, the use of the metastatic variable serves as a proxy cancer stage variable as has been used by previous studies using the Healthcare Cost and Utilization Project–NIS database.35,52 Second, we used in-hospital mortality as a proxy for assessing clinical outcomes for patients with HNC. A better approach would have been to examine 30-day, 60-day, or 90-day mortality after hospitalization for cancer. The NIS does not capture long-term mortality or detailed follow-up clinical information. It is likely that uninsured or inadequately insured patients would have had even worse long-term mortality rates because they may not be able to afford postdischarge care. Therefore, the estimated, risk-adjusted, in-hospital mortality rates in the current study may have underestimated the true mortality rates for uninsured patients. Moreover, our nomogram had 72% accuracy; therefore, other large representative data sets with information on biomarkers and clinical variables such as tumor size are needed to improve predictive accuracy.
Our study examined multilevel factors associated with in-hospital mortality of patients with HNC. We found that age, sex, LOS, type of admission, admission day, and number of comorbidities were independent patient-level predictors of in-hospital mortality and that hospital teaching status was an independent hospital-level predictor of in-hospital mortality. We developed a nomogram to be used as a clinical instrument for hospitalized patients with HNC to estimate their risk of in-hospital mortality. This information can help identify patients at higher risk of in-hospital mortality in the future. Greater levels of care and surveillance may improve their hospital course.
Corresponding Author: Eric Adjei Boakye, PhD, Department of Population Science and Policy, Southern Illinois University School of Medicine, 201 E Madison St, PO Box 19664, Springfield, IL 62794 (email@example.com).
Accepted for Publication: January 15, 2020.
Published Online: March 19, 2020. doi:10.1001/jamaoto.2020.0132
Author Contributions: Drs Adjei Boakye and Osazuwa-Peters had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Concept and design: Adjei Boakye, Osazuwa-Peters, Chen, Challapalli, Piccirillo.
Acquisition, analysis, or interpretation of data: Adjei Boakye, Osazuwa-Peters, Chen, Cai, Tobo, Buchanan.
Drafting of the manuscript: Adjei Boakye, Osazuwa-Peters, Chen, Tobo, Challapalli.
Critical revision of the manuscript for important intellectual content: Adjei Boakye, Osazuwa-Peters, Chen, Cai, Buchanan, Piccirillo.
Statistical analysis: Adjei Boakye, Cai, Tobo.
Administrative, technical, or material support: Chen.
Supervision: Adjei Boakye, Chen, Buchanan, Piccirillo.
Conflict of Interest Disclosures: None reported.
Meeting Presentation: The abstract of this study was presented at the North American Association of Central Cancer Registries 2016 Annual Conference; June 14, 2016; St Louis, Missouri.
Create a personal account or sign in to: