Development and Validation of the Neck Dissection Impairment Index: A Quality of Life Measure

Objectives  To validate a health-related quality-of-life (QOL) instrument for patients following neck dissection and to identify the factors that affect QOL following neck dissection.

Design  Cross-sectional validation study.

Setting  The outpatient clinic of a tertiary care cancer center.

Patients  Convenience sample of 54 patients previously treated for head and neck cancer who underwent a selective neck dissection or modified radical neck dissection (64 total neck dissections). Patients had a minimum postoperative convalescence of 11 months. Thirty-two underwent accessory nerve–sparing modified radical neck dissection, and 32 underwent selective neck dissection.

Main Outcome Measure  A 10-item, self-report instrument, the Neck Dissection Impairment Index (NDII), was developed and validated. Reliability was evaluated with test-retest correlation and internal consistency using the Cronbach α coefficient. Convergent validity was assessed using the 36-Item Short-Form Health Survey (SF-36) and the Constant Shoulder Scale, a shoulder function test. Multiple variable regression was used to determine variables that most affected QOL following neck dissection

Results  The 10-item NDII test-retest correlation was 0.91 (P<.001) with an internal consistency Cronbach α coefficient of .95. The NDII correlated with the Constant Shoulder Scale (r = 0.85, P<.001) and with the SF-36 physical functioning (r = 0.50, P<.001) and role–physical functioning (r = 0.60, P<.001) domains. Using multiple variable regression, the variables that contributed most to QOL score were patient's age and weight, radiation treatment, and neck dissection type.

Conclusions  The NDII is a valid, reliable instrument for assessing neck dissection impairment. Patient's age, weight, radiation treatment, and neck dissection type were important factors that affect QOL following neck dissection.

WE EVALUATED patients with head and neck cancer to assess the long-term effects of neck dissection on quality of life (QOL) related to shoulder dysfunction. Patients frequently have varying degrees of shoulder impairment following neck dissection, which range from mild to profound dysfunction.^1-4 What is not well established is whether the degree of shoulder impairment observed has significant QOL implications in this population. Our objective was to determine the specific complaints and difficulties that patients report after neck dissection and to design a reliable and valid self-report instrument that would enable us to assess the QOL impact related to neck dissection.

The shoulder syndrome, characterized by Nahum,⁵ is a constellation of symptoms that include shoulder pain, limitations of abduction, and scapular winging. Depending on the extent of the neck dissection, patients can have a variable amount of difficulty with shoulder syndrome. Even after convalescence, the shoulder impairment can be lasting and substantial enough to affect an individual's QOL through changes in employment status, recreational pursuits, and personal independence. When serious impairment occurs, major changes in lifestyle are necessary to accommodate patients' limitations.^6-8

To evaluate the factors that affect QOL after neck dissection, we developed a self-administered questionnaire, the Neck Dissection Impairment Index (NDII). The problems related to this population are unique and require investigation specific to their issues.⁹ Standard psychometric methods were used to evaluate the validity and reliability of the instrument.^10-12 It was our goal to develop a questionnaire that was reliable and easy to administer and that would evaluate the difficulties patients have following neck dissection.

We conducted a cross-sectional study to evaluate the factors that affect patients following neck dissection procedures. Our intention was to design and validate a self-administered survey questionnaire that measured patient's QOL related to shoulder impairment.

On study entry, patients received the NDII and the Medical Outcomes Study 36-Item Short-Form Health Survey (SF-36) following informed consent. Permission to use the SF-36 was approved by the Medical Outcomes Trust. Subsequently, each patient performed the Constant shoulder functional test after self-administration of the 2 questionnaires. For patients who had undergone bilateral neck dissections, each side was evaluated and scored separately with the NDII and the Constant Shoulder Scale.

There were 54 patients in the study group who had undergone a total of 64 neck dissection procedures (10 patients received bilateral procedures). A total of 32 accessory nerve–sparing modified radical neck dissections (MRNDs) and 32 selective neck dissection (SND) procedures were performed in this group. The mean ± SD age of the study population was 56.8 ± 11.7 years, and the mean weight was 73.5 ± 14.9 kg. Thirty-nine patients were male; 15 were female; and the mean time elapsed from surgery was 33.7 months (range, 11-120 months). Tumors were found in the oral cavity in 10 patients; oropharynx, 23 patients; larynx or hypopharynx, 10 patients; other sites, 8 patients; and unknown primary sites, 3 patients. Tumors were staged according to the TNM staging system (according to the AJCC Cancer Staging Manual) as TX for 4 patients, T1 through T2 for 15 patients, T3 through T4 for 35 patients, N0 for 30 patients, N1 for 10 patients, N2 for 23 patients, and N3 for 1 patient. Most patients received both surgery and radiation therapy (87%), and squamous cell carcinoma was the most common tumor type (92.5%).

For the development and validation of the questionnaire, patient inclusion criteria were having previously untreated and diagnosed head and neck cancer and concurrently requiring selective or modified neck dissection as part of the management of head and neck cancer. Patients were excluded if they had undergone surgery less than 11 months previously, reported any history of unrelated neck or shoulder pathologic conditions, or had known recurrent disease at the time of evaluation. Enrolled patients received their head and neck cancer treatment at the Department of Otolaryngology, University of Michigan, Ann Arbor, and were encountered on an outpatient basis during their routine follow-up care. The University of Michigan's Institutional Review Board approved study materials and methods.

Initial topics selected for the questionnaire were developed after a review of the relevant literature concerning shoulder dysfunction after neck dissection.^1-8 Additional topics were selected after interviews with approximately 40 patients in an outpatient setting who had previously undergone a neck dissection procedure. Patients responded to open-ended questions that pertained to topics that included (but were not limited to) neck and shoulder symptoms, limitations in activities of daily living, occupational and leisure activities, and social and emotional effects related to shoulder impairment. Survey items were refined by a panel that consisted of otolaryngologists, physical therapists specializing in rehabilitation of postsurgical oncology patients, and survey specialists from the University of Michigan School of Public Health. Finally, the items were pilot-tested on a group of 25 patients undergoing neck dissections to further evaluate survey comprehension, content, and clarity. During patient piloting, it was clear that patients who had undergone bilateral procedures could readily distinguish between symptoms and limitations that emanated from one side vs the other. Patient instructions designated that responses to each item be considered "during the last 4 weeks." For each survey item, a Likert scale was used with 5 response options. Initially, the NDII contained 15 items; 10 items remained after item reduction. Figure 1 shows the 10 items (in summary form) that remained after validation. Scoring was achieved by rating response items from 1 to 5, with 5 representing better QOL related to neck dissection. Scores were then transformed to a 0 to 100–point scale.

The SF-36 is a validated and widely used general health QOL survey that assesses 8 areas (domains) of general health.^13,14 The 8 areas relate to physical functioning, role–physical functioning, bodily pain, role–emotional, social functioning, mental health, vitality, and general health perceptions. Each domain is scored from 0 to 100, with higher scores representing better health QOL. We selected the SF-36 to establish concurrent validity with the NDII because of the extensive use of the SF-36 across many patient groups as a general measurement of health-related QOL.

The Constant Shoulder Scale was also selected to support criterion validity. It is a validated clinical assessment of shoulder function that has established utility and accuracy across all diseases that affect the shoulder. This validated, widely used clinical test is an accurate and sensitive measure of shoulder function, detecting subtle changes in shoulder function.¹⁵ It is a weighted test that combines patient symptom scores and objective measures of active shoulder range of motion, combined internal rotation, combined external rotation, and shoulder strength in the plane of the scapula. Scores range from 0 to 100, with higher scores indicating better shoulder function. Demonstrating a strong correlation between this functional measurement and a measure of QOL will provide support that QOL is affected by shoulder function.

All patient data, including demographic information, survey data, and disease information, were compiled in a relational database on a personal computer. Data entries were checked and verified against the primary data. All statistical procedures were performed using statistical software (SAS version 6.12; SAS Institute Inc, Cary, NC). Both a principal components analysis and an exploratory factor analysis with a varimax transformation were initially performed on the NDII. Reliability was established in 2 manners: single-item and total score test-retest correlation and internal consistency with Cronbach α. Both the factor analysis and the reliability measures were used to determine item reduction. Items with less than 0.50 retest correlation were not considered reliable and thus not included in the final index. Content and face validity were confirmed with patient piloting and interviews, whereas discriminant validity was evaluated by exploring correlation with SF-36 domains and the Constant Shoulder Scale.

Multivariable regression analysis was used to model the dependent variables. Univariate and stepwise regression analyses were used to facilitate selection of variables. Preliminary analysis of variables included evaluation for the presence of multicollinearity, such that candidate models contained no violations of multicollinearity. Independent variables assessed included patient age, time elapsed from surgery, tumor stage (T1-T2 or T3-T4), tumor site (oral cavity, oropharynx, and hypopharynx and larynx), patient weight, radiation therapy, handedness, and neck dissection type (SND or MRND).

Exploratory factor analysis with varimax transformation revealed 1 large factor and 2 smaller factors; factors with eigenvalues less than 1 were not considered. A 10-item factor that addressed physical abilities and activities was identified as the main domain; a second factor without an apparent unifying theme consisted of items relating to driving, appearance, and sleep. The third factor had a 2-item cluster, which related to stress and QOL. Individual item and factor eigenvalues are listed in Table 1. The 10-item factor was ultimately used for the final version of the NDII.

Both individual item and total score test-retest correlation were performed using the Spearman log-rank correlation to assess reliability. Respondents were asked to complete the questionnaire 5 days after initial entry; 87% of patients who were retested completed the survey. Total score correlation was 0.91, whereas individual item correlation ranged from 0.41 to 1.00.

Internal consistency was determined using the Cronbach α coefficient. The initial 15-item survey had an overall α coefficient of .94; however, 5 items had poor test-retest reliability coefficients or poor internal consistency and therefore were omitted in the final 10-item version of the NDII. The omitted items correspond to the items in the smaller factors 2 and 3. The remaining 10 items contained within factor 1 were used for further analysis. The 10-item version had an overall internal consistency of .95.

Convergent validity was evaluated using the Constant Shoulder Scale. The mean (SD) Constant Shoulder Scale score for all patients was 70.7 (17.4) (range, 38-100). Our hypothesis was that the NDII would have a high, positive correlation with this instrument. The overall correlation between the 2 was 0.85 (P<.001). Correlating the NDII with the SF-36 domains that were hypothesized to correlate with it also tested convergent validity. Specifically, we hypothesized that the role–physical, physical functioning, and body pain domains would have good correlation with the NDII. The SF-36 role–physical (r = 0.60, P<.001) and physical function domains (r = 0.50, P<.001) had good correlation with the NDII survey, whereas the body pain domain had relatively weak but significant correlation (r = 0.32, P = .005) (Table 2). The social functioning and mental health SF-36 domains also exhibited good correlation with the NDII.

Individual items from the 10-question NDII were scored from 1 to 5, with 5 representing higher QOL responses. The total NDII score was then scaled to a 100-point cumulative score. The overall mean ± SD score for all patients was 67.8 ± 17.4 (range, 7.5-100.0). Figure 2 demonstrates mean scores (1 to 5) for each item question.

After validating the NDII, we used the data collected from the NDII and SF-36 to determine the factors that best predicted long-term shoulder QOL. We used the NDII and SF-36 as dependent variables for QOL in multivariable regression analyses and factored in important treatment, clinical, and demographic variables. When analyzing the SF-36, we evaluated total composite score, individual domains, and summary scores from the physical component summary and mental component summary. The best models for predicting QOL-related shoulder function were obtained with the NDII as the dependent variable. This supports the NDII as a useful disease-specific measure of shoulder-related QOL. No single domain or summary score of the SF-36 could fit as good a model. Of the SF-36 domains and summary scores, the physical component summary scale was the best QOL predictor (R² = 0.32, P<.001).

The best NDII model predicting shoulder-related QOL included the following parameters: patient age in years (P = .12), weight in kilograms (P<.001), neck dissection type (P = .14), and use of external beam radiation therapy (P = .04). The final model equation was as follows:

NDII = 5.44 + 0.82 (Kilograms) − 13.45 (Radiation Therapy) + 0.30 (Age) − 6.43 (Neck Dissection Type).

Several factors led to the selection of this model. Of the candidate models, this model had one of the best R² values (0.44), had a well-matched C_p criterion, and was significant (P<.001). The variables contained in this model were each found in the best 1-, 2-, and 3-parameter models obtained with stepwise multivariable regression analysis. Radiation was used on 87% of our sample population. We evaluated whether including radiation therapy in the model would serve as a confounder or surrogate for other clinical and demographic factors, such as extent of disease or type of neck dissection performed. Therefore, the issues of multicollinearity and bias of radiated versus nonradiated patients within the other variables were carefully examined. We found no violations of multicollinearity, and radiation therapy independently provided a substantial magnitude to the overall model equation. This 4-variable model was chosen over models with more variables because more variables did not appreciably strengthen the model. To understand the magnitude of contribution of each of the variables to the overall model, the following is an example of how the model equation can be interpreted. In a 70-year-old, 70-kg patient, weight would contribute nearly 55 points (0.82 × 70 kg) to the overall score (maximum of 100 points), and age would contribute 21 points (0.30 × 70 years). However, the addition of radiation would decrease the overall score by 13.45 points, and MRND would decrease the score by 6.43 points.

In this study we were able to design and validate a QOL-specific instrument in patients who had undergone neck dissection and correlate long-term impairment with the Constant Shoulder Scale, a validated shoulder function assessment. Excellent reliability and validity of the NDII was demonstrated by test-retest correlation (r = 0.91), internal consistency (r = 0.95), and good convergent validity with the Constant Shoulder Scale (r = 0.85) and SF-36 domains (role–physical and physical functioning domains). Furthermore, when the SF-36 was modeled as the dependent variable, no multivariable regression model using the SF-36, or any of its domains and component scales, was as good as the NDII.

The variables that most affected long-term shoulder impairment were identified through multiple variable regression analysis. These factors included the type of neck dissection performed (SND vs MRND), treatment with external beam radiation, patient's age, and patient's weight.

The type of neck dissection performed was an important factor; patients treated with MRND had worse NDII scores than those treated with SND. Other studies have evaluated shoulder function following different types of neck dissection procedures. Remmler et al⁸ serially compared accessory nerve–sparing procedures with nerve-sacrificing procedures (eg, radical neck dissection) using a battery of physical therapy measurements. They found that nerve–sparing procedures were associated with better preservation of shoulder function. Sobol and Jensen¹ found that patients undergoing SND demonstrated functional capacity and perception of shoulder pain 16 weeks following surgery that was superior to those undergoing MND and radical neck dissection; however, there were limited long-term data available in their study. Also, none of the aforementioned studies used a validated QOL instrument. Recently, Terrell et al¹⁶ reported findings from the University of Michigan Head and Neck QOL (UMHNQOL), an instrument that specifically assessed shoulder pain as a single item within the context of a general head and neck QOL instrument. They reported that patients undergoing nerve-sparing level V neck dissections experienced more shoulder or neck pain than patients in whom level V was undissected. Compared with the UMHNQOL, the NDII assesses shoulder-related QOL in a more comprehensive and detailed manner. It is a multi-item, validated instrument developed to evaluate a wide range of components of shoulder-related QOL after neck dissection. An analysis of single items in the NDII showed significant QOL differences in pain scores between patients undergoing SND and those undergoing MRND. Additionally, the NDII demonstrated that there were also significant QOL differences by neck dissection type in the areas of lifting light objects and leisure and recreation pursuits.

Radiation treatment was a significant variable that was a predictor for unfavorable shoulder-related QOL in our study. Schuller et al⁶ sought to evaluate the impact of surgery and radiation on shoulder function. The study found that the patients who received radiation in addition to surgery more frequently reported an increased reliance on others. Although an excellent study with important findings, a validated QOL instrument needed to be used to evaluate this patient population. Our study confirmed the results of the study by Schuller and colleagues with a validated questionnaire (NDII) and demonstrated that the use of radiation in patients undergoing neck dissections is an independent, negative prognosticator of shoulder function.

Weight was not only important to the model, it was also significant as an independent variable. We postulated that weight was a surrogate for overall patient physical well-being and nutritional status and therefore had a positive correlation with health-related QOL for shoulder function. It seems intuitive that more robust patients would have better recuperative and compensatory powers for maintenance of shoulder function and, thus, have less QOL-related impairment.

Age was an important contributor in our regression model for shoulder-related QOL following neck dissection, even after controlling for the other clinical-demographic factors. The older the patient, the less impact there was in QOL. It was our supposition that older patients might not perceive substantial functional deficits as having proportionally as great an impact on QOL, perhaps due to lower demands on shoulder function or lower expectations of their physical capacity related to their shoulder function compared with younger patients.

It was our goal to develop and validate a QOL instrument to assess shoulder function and to systematically determine which treatment, clinical, and demographic factors best predict long-term shoulder QOL impairment following neck dissection. Thus, we developed a validated QOL instrument, the NDII. Age, weight, radiation treatment, and neck dissection type were the variables that most affected QOL. Higher-risk patients (eg, younger, less robust patients who receive radiation treatment and MRND) might benefit from early, more aggressive rehabilitative intervention.

Accepted for publication August 6, 2001.

Corresponding author and reprints: Douglas B. Chepeha, MD, MSPH, Department of Otolaryngology–Head and Neck Surgery, 1500 E Medical Center Dr, Taubman Center 1904, University of Michigan, Ann Arbor, MI 48109 (e-mail: dchepeha@umich.edu).