Annals of Oncology Advance Access originally published online on October 19, 2005
Annals of Oncology 2006 17(1):151-156; doi:10.1093/annonc/mdj020
© 2005 European Society for Medical Oncology
Selecting predictors of cancer patients' overall perceptions of the quality of care received
Department of Health Policy, Management, and Evaluation, University of Toronto, Toronto, Ontario, Canada
* Correspondence to: Dr A. D. Brown, 150 College Street, Room 147D, Toronto, Ontario, Canada M5S 3E2. Tel: +1-416-978-1484; Fax: +1-416-978-1466; E-mail: adalsteinn.brown{at}utoronto.ca
 |
Abstract |
|---|
 Top Abstract introductionpatients and methodsresultsdiscussionReferences |
|---|
Background: The goal of this study was to identify aspects of care (predictors) that can most easily be modified to produce an improvement in the score of patients' overall evaluations of the quality of care received.
Patients and methods: Our sample consisted of 2247 cancer patients hospitalized in Ontario acute care hospitals in 1999/2000. We sought predictors of patients' overall perceptions of the quality of care by applying a methodology that minimizes the improvement of the predictors while gaining a desired increase in the score of the dependent variable. This approach tends to ignore items that rate relatively high and focuses on those for which hospitals can more easily modify the score. Two main subgroups were analyzed in this study: patients with malignant and benign neoplasms.
Results: ‘Skills of nursing staff’, ‘courtesy of nursing staff’, ‘courtesy of people who drew blood’ and ‘cleanliness of hospital in general’ were consistently selected by our approach in both cancer groups.
Conclusions: This study identifies an efficient approach to improving the score of patients' overall perceptions of the quality of care received. By focusing on these aspects of care, hospitals may be able to improve the allocation of scarce resources when planning patient satisfaction improvement initiatives.
Key words: cancer patients, patient satisfaction, perceptions of the quality of care, predictors
|
introduction |
|---|
|
TopAbstractintroductionpatients and methodsresultsdiscussionReferences |
|---|
Patient satisfaction has become an important performance measure in the last decade. Although it has been argued that patient satisfaction is not always clearly defined [1
, 2] or well understood [3], it is considered an important aspect of measuring the quality of the care process [3–7]. Literature addressing patient satisfaction in oncology settings has gained more attention lately, but little is known about what most influences patients' overall assessments of the care received. Using different approaches, some studies have addressed this question [8–12], and it appears that aspects of care related to nurses and doctors are among those that most influence cancer patients' overall appraisals of their health-care experience. Interventions aimed at improving patient satisfaction may become of great significance since there is evidence to suggest that satisfied patients are more likely to cooperate with their treatment [1, 3, 5, 8, 10, 12, 13], which is, in turn, associated with better clinical outcomes [12]. These relationships are even more important for patients with chronic illness such as cancer [12].
From a hospital's perspective, one of the main reasons for measuring patient satisfaction is to provide information to facilitate improvements to the process of care. This study aimed to identify aspects of the care process that can most easily be modified to yield a desired level of improvement in the score of a variable that measures cancer patients' overall perceptions of the quality of care received. To reach this goal, we applied a methodology that combines the use of regression and optimization models [14]. The value of this approach is that it considers the strength of the predictors (aspects of care), as well as their current value or performance. As a result, aspects of care selected by this approach tend to be those that score relatively low but are highly associated with the variable that measures patients' overall appraisals of their health-care experience. In other words, it may require more effort to improve a patient satisfaction score that is very high than to modify one that is lower. The optimization technique addresses this issue by ignoring the items that rate relatively high and focuses on those for which hospitals can more easily modify the score.
From a management perspective, our findings might help health-care providers allocate scarce resources more effectively when planning patient satisfaction improvement initiatives. The Parkside instrument (Parkside Associates Inc.) validated by Carey and Seibert [6] in 1990 in the USA was used in this study to measure patient satisfaction.
|
patients and methods |
|---|
|
TopAbstractintroductionpatients and methodsresultsdiscussionReferences |
|---|
sampling
A database containing results from an inpatient satisfaction survey and discharge information for 26 276 patients was obtained from the Hospital Report Project (HRP) [15
]. These patients visited 104 Ontario acute care hospitals for general medical or surgical care between April of 1999 and September of 2000 and had an overnight stay of one or more days. From this database, we extracted 2275 patients who had a malignant or benign neoplasm code as the ‘most responsible diagnosis’, meaning the condition most responsible for the patient's length of stay in hospital. After excluding 28 patients who had fewer than 20 of the 66 survey questions answered, our final sample included 2247 cancer patients who were hospitalized in one of 99 acute care hospitals in Ontario, Canada (teaching = 13, community = 71 and small = 15). Table 1 shows detailed information on the sample.
|
the questionnaire
Originally, the HRP mailed satisfaction surveys between October and December 2000 to 74 926 patients who stayed overnight in Ontario acute care hospitals. The survey is a modified version of an inpatient satisfaction questionnaire developed by Parkside Associates Inc. It was psychometrically validated by Carey and Seibert [6
], who surveyed close to 40 000 inpatients from over 196 hospitals who were discharged during May 1990 in the USA. Their overall response rate of 44% is fairly close to the 40.5% reported by the HRP. The modified HRP survey includes 66 closed-ended items plus two open-ended items. Two main scales are used in the survey: a 6-point scale (1 = excellent, 2 = good, 3 = fair, 4 = poor, 5 = very poor, 6 = does not apply) and a 4-point scale (1 = yes, 2 = to some extent, 3 = no, 4 = does not apply). The aspects of care evaluated in the survey include items related to overall ratings, admission processes, nursing care, physician care, care by other providers, hospital care (e.g. pain control, medication management, communication, etc.), preparation for discharge, housekeeping, self-rated health status and prior health service use [15].
statistical analysis
Similar to Carey and Seibert [6], we selected a single item from the patient satisfaction questionnaire as the dependent variable, which was ‘what is your overall opinion of the quality of care received?’ This single-item measure is also similar to other studies [9, 11, 12], and it has been argued that this approach is more practical than the use of longer or multi-item measures [12].
The independent variables (aspects of care) were also single items from the questionnaire, and were selected based on the following criteria. First, from a practical perspective, questions that were too broad to allow hospitals to design specific improvement actions were excluded (for example, ‘what is your overall opinion of nursing care?’). Secondly, questions not related to aspects of care such as ‘was this your first time at this hospital?’ were also excluded. Finally, we also excluded survey questions with a high percentage of responses missing or those with responses of ‘does not apply’. Although there is no rigid rule to determine which questions are to be excluded because of missing values, we used a guideline of 15% or more. As a result, our independent variables included 34 survey questions (out of the original 66). All of them were significantly associated with the dependent variable (Spearman's r range 0.10–0.60; P < 0.0001) and related to physician care, nursing care, care by other providers, admission waiting time, housekeeping, preparation for discharge and hospital care. Again, our single-item approach for the independent variables is similar to other studies [9, 11, 12], and appears to be more practical when planning patient satisfaction improvement initiatives [16].
We simultaneously entered all 34 independent variables into an ordinal logistic regression model to explain the variance of patients' overall perceptions of the quality of care received (the dependent variable). Two regressions were performed: (i) patients with malignant neoplasm (n = 1829); and (ii) patients with benign neoplasm (n = 418). Missing values and responses of ‘does not apply’ for each patient were replaced using Monte Carlo Markov Chain methods [17]. All items were also transformed to a 5-point scale. Non-parametric Kruskal–Wallis and Wilcoxon two-Sample tests were also performed to explore differences in patients' overall judgments of the quality of care received. Data management and analyses were performed using The SAS System for Windows, version 8.02 (SAS Institute Inc., Cary, NC, USA).
the optimization model
The model below aims to select predictors (aspects of care) that, for the least amount of improvement, yield a desired increase in the score of the variable that measures patients' overall perceptions of the quality of care received. This model selects predictors based on both their current performance and their strength (measured by the magnitude of the multiple regression coefficients).
Minimize (model originally formulated by Brown et al. [14]):
 |
 |
 |
 |
 |
Where Xi represents the optimization variables (predictors) and Yi reflects the current average performance of the independent variables (predictors). For example, X2 denotes the question ‘how would you rate the courtesy of the nursing staff?’, which has a current average performance of 1.47 (Y2) for patients with malignant neoplasm on a scale of 1 to 5 (1 = excellent and 5 = very poor). Restriction 4 was relaxed to allow the model to select up to five predictors. Thus, we preset restriction 4 to 11% for the model involving patients with malignant cancer, and to 16% for the model involving patients with benign neoplasm. Microsoft Excel Solver (Microsoft Corporation, Redmond, WA, USA) was used to find optimal solutions that met the specified conditions. The optimization algorithm was the generalized reduced gradient method [18].
|
results |
|---|
|
TopAbstractintroductionpatients and methodsresultsdiscussionReferences |
|---|
patients' overall perceptions of the quality of care received
A total of 88% of the cancer patients (both malignant and benign) judged the quality of care as excellent or good (Figure 1). We found no statistically significant differences in patients' overall perceptions of the quality of care between those with malignant and benign neoplasm. However, we found that patients hospitalized once during the past 2 years evaluated significantly higher the quality of care than those hospitalized three and four times (P < 0.05). We also found that less healthy cancer patients (self-assessed health) tended to judge the quality of care lower than healthier cancer patients (P < 0.0001). Patients' overall perceptions of the quality of care were also lower in cases where someone other than the patient completed the survey (P < 0.0001). Finally, we found no statistically significant associations between the dependent variable and age, gender or length of stay.
|
results from the optimization model
For patients with malignant cancer, the five predictors selected by the optimization model to increase the score of the dependent variable by 11% were: ‘courtesy of nursing staff’, ‘courtesy of people who drew blood’, ‘courtesy of people who delivered food’, ‘skills of nursing staff’ and ‘cleanliness of hospital in general’. For patients with benign neoplasms, the five predictors selected to increase the score of the dependent variable by 16% were: ‘courtesy of nursing staff’, ‘courtesy of people who drew blood’, ‘skills of nursing staff’, ‘cleanliness of room’ and ‘cleanliness of hospital in general’. The model also suggested that the score of most of these predictors needs to be improved by 15% to gain the desired increase in the score of the dependent variable (Table 2). Figure 2 also shows the number of predictors necessary to increase the score of the dependent variable with restriction 4 ranging from 1% to 21%.
|
|
results from the regression model
Table 3 shows the ordinal logistic regression models for patients' overall perceptions of the quality of care received. For patients with malignant cancer, the five predictors that most explained the variance of the dependent variable (measured by the magnitude of the standardized multiple regression coefficients) were: ‘skills of nursing staff’, ‘courtesy of nursing staff’, ‘cleanliness of hospital in general’, ‘thoroughness of nursing staff’ and ‘courtesy of people who drew blood’. The model for this patient group explained 56% of the variance, and the five strongest predictors explained 50%. For patients with benign neoplasms, the five predictors that most positively explained the variance of the dependent variable were: ‘cleanliness of hospital in general’, ‘courtesy of people who drew blood’, ‘cleanliness of room’, ‘courtesy of nursing staff’ and ‘skills of nursing staff’. The model for this patient group explained 64% of the variance, and the previous five predictors explained 52%.
|
|
discussion |
|---|
|
TopAbstractintroductionpatients and methodsresultsdiscussionReferences |
|---|
This study focused on identifying efficiently aspects of care (predictors) that need to be modified to produce an improvement in the score of patients' overall perceptions of the quality of care received (the dependent variable). Overall, our approach favored predictors with large and positive regression coefficients (ß in Table 3) and those for which their current performance was relatively low, since they produced a greater effect on the score of the dependent variable. As a result, predictors selected were not always the strongest identified through regression models. Our technique is fairly similar to that of Gesell and Gregory [19
], who prioritized service areas to implement patient satisfaction improvement programs for cancer patients. Some of their priority aspects of care selected were not among those that most correlated to their overall satisfaction measure. In practice, their approach and ours suggest that looking merely for the strongest predictors may not always be the most efficient way to influence patients' overall evaluations of the care received. A similar rationale was also used by Gourdji et al. [20], who combined satisfaction rates and importance rates (instead of correlations or regression coefficients) to identify service areas for patient satisfaction improvement efforts.
The score for patients' overall perceptions of the quality of care (88%) is high, but it is similar to other studies [8–12, 19, 20]. We also found no differences in patients' overall perceptions of the quality of care among those of different sex or age, although there is some evidence to support the fact that older patients often report greater satisfaction rates than younger patients [9, 12]. Different from Skarstein et al. [8], we found that patients hospitalized once in the previous 2 years tended to judge the quality of care higher than those hospitalized three or four times. The lack of a relationship between the dependent variable and length of stay is also in agreement with Pascoe [3], who concluded that the duration of a patient–provider visit is not consistently related to patient satisfaction.
For both patient groups (malignant and benign), the only technical aspect of care selected by the optimization model was ‘skills of nursing staff’. This is not surprising in light of other studies that have positively associated providers' technical competence with patient satisfaction [3, 8, 10, 14]. Interestingly, the aspect of care most frequently selected by our approach in both patient groups relates to courtesy of caregivers (Table 2). Again, this finding is consistent with many others who have identified interpersonal skills as a major determinant of patient satisfaction [3, 6, 11, 14, 21]. Gesell and Gregory [19] also suggested that improvements in service quality and patients' perceptions of service can be achieved through the modification of the care providers' behavior and attitude.
‘Courtesy of physicians’, the fifth strongest predictor for patients with malignant neoplasm, was avoided by the optimization algorithm since it scored relatively high. This appeared to be similar to Gourdji et al. [20], who found many physician aspects of care with high importance rates. However, satisfaction rates were also high, and thus they were not among the priority areas for patient satisfaction improvement purposes. All other physician aspects of care in our study were weakly associated with the dependent variable; thus, they were less likely to be selected by the optimization algorithm.
‘Acceptable waiting time to be admitted’ was not among the strongest predictors in the regression models, nor was it selected by the optimization algorithm. This is similar to other studies that did not find aspects of care related to waiting times as relevant among cancer patients [6, 8–10, 12]. However, Shilling et al. [22] found that satisfaction with the length of wait in clinic was strongly associated with overall patient satisfaction. Similarly, none of the predictors selected by our optimization model were related to level of information received or caregiver-patient communication. This is different from other studies that found these aspects of care to be highly associated with patient satisfaction [8, 9, 11].
We further analyzed a subsample of 1184 cancer patients from the original 2247. This group of patients were those with the most common cancer diagnoses: colon, rectum, lung, female breast, prostate, bladder and benign uterine leiomyoma (Table 1). The aspects of care selected by our approach for this patient group were ‘courtesy of nursing staff’, ‘courtesy of people who drew blood’, ‘courtesy of people who delivered food’, ‘skills of nursing staff’ and ‘cleanliness of hospital in general’. These predictors are similar to those selected for patients with malignant and benign neoplasm (the main analyses in this study), suggesting that strategies aimed at improving patient satisfaction scores may be similar regardless of the type of cancer. Larger sample sizes would be required to develop valid regression models to test predictors specific to individual cancer types, such as colon cancer.
We also further tested our approach in a scenario in which patients' perceptions of their health-care experience were relatively low. We divided the data in half, with the first half representing those hospitals performing relatively high on patient perceptions (65 hospitals; 1118 patients), and the second half those performing relatively low (34 hospitals; 1129 patients). In both cohorts, our approach selected predictors related to courtesy of caregivers, skills of nursing staff and cleanliness of the facility. These predictors are similar to those selected for patients with malignant and benign neoplasm (the main analyses in this study). While this further analysis resulted in the identification of some additional predictors for both the high and low performing groups of hospitals, they were not sufficiently strong to make a difference in the selection of key predictors linked to patients' overall perceptions of their care experience.
Our approach tended to select a smaller number of predictors when the score of the dependent variable was lower. Figure 2 shows that, regardless of the preset level of increase in the score of the dependent variable, the number of required predictors is consistently smaller for patients with benign neoplasms, who had slightly lower scores for the dependent variable. In practice, this means that improving the score of patients' overall perceptions of the quality of care among those with benign neoplasms may require less effort than improving the score of this overall appraisal among patients with malignant cancer.
Predictors selected by the optimization algorithm were similar to those that best explained the variance of the dependent variable. Although the findings obtained from the optimization approach may appear similar to those one would identify from the regression technique, there are some differences. First, the optimization model not only identifies the predictors to focus on, but it also provides the percentage improvement required by the predictors to gain a desired increase in the score of the dependent variable. Secondly, by considering the current performance of the predictors, the optimization algorithm tends to avoid those that score relatively high. Finally, the predictors from the optimization algorithm were selected under a criterion which specified minimization of the total improvement of the predictors. This criterion is important from a management perspective as it allows managers and clinicians to focus initiatives linked to predictors that strongly influence patients' overall perceptions and that may require less effort or resources to impact these overall appraisals.
A potential limitation to our study is that we did not include a cost structure in the formulation of the optimization model. Consequently, implementing intervention strategies based on the results of this study may not result in the most cost-effective solution. Our model assumes that the cost to increase the score of each predictor is equivalent. Further investigation is required to explore costs related to improving the score of the predictors of patients' overall perceptions of the quality of care received.
|
Acknowledgements |
|---|
This work was supported in part by the Hospital Report Project, a joint initiative of the Government of Ontario and the Ontario Hospital Association.
Received for publication May 24, 2005. Revision received August 4, 2005. Accepted for publication August 12, 2005.
|
References |
|---|
|
TopAbstractintroductionpatients and methodsresultsdiscussionReferences |
|---|
1. Strasser S, Aharony L, Greenberger D. The patient satisfaction process: moving toward a comprehensive model. Med Care Rev 1993; 50: 219–248.[Medline]
2. O'Connor SJ, Shewchuk R. Patient satisfaction: what is the point? Health Care Manage Rev 2003; 28: 21–24.[Web of Science][Medline]
3. Pascoe GC. Patient satisfaction in primary health care: a literature review and analysis. Eval Program Plann 1983; 6: 185–210.[CrossRef][Medline]
4. Donabedian A. The quality of care. How can it be assessed? JAMA 1988; 260: 1743–1748.
5. Larsson BW, Larsson G, Chantereau MW, von Holstein KS. International comparisons of patients' views on quality of care. Int J Health Care Qual Assur 2005; 18: 62–73.
6. Carey RG, Seibert JH. A patient survey system to measure quality improvement: questionnaire reliability and validity. Med Care 1993; 31: 834–845.[Web of Science][Medline]
7. Vardy J, Tannock IF. Quality of cancer care. Ann Oncol 2004; 15: 1001–1006.
8. Skarstein J, Dahl AA, Laading J, Fosså SD. Patient satisfaction' in hospitalized cancer patients. Acta Oncol 2002; 41: 639–645.[CrossRef][Web of Science][Medline]
9. Blanchard CG, Labrecque MS, Ruckdeschel JC, Blanchard EB. Physician behaviors, patient perceptions, and patient characteristics as predictors of satisfaction of hospitalized adult cancer patients. Cancer 1990; 65: 186–192.[CrossRef][Web of Science][Medline]
10. Kavadas V, Barham CP, Finch-Jones MD et al. Assessment of satisfaction with care alter inpatient treatment for oesophageal and gastric cancer. Br J Surg 2004; 91: 719–723.[CrossRef][Web of Science][Medline]
11. Fosså SD, Hjermstad MJ, Mork IH, Hjortdahl P. Does the service at a large oncologic out-patient clinic satisfy the patients' perceived need? Int J Health Care Qual Assur 1996; 9: 24–29.[Medline]
12. Walker MS, Ristvedt SL, Haughey BH. Patient care in multidisciplinary cancer clinics: does attention to psychosocial needs predict patient satisfaction? Psychooncology 2003; 12: 291–300.[CrossRef][Medline]
13. Fakhoury W, McCarthy M, Addington-Hall J. Determinants of informal caregivers' satisfaction with services for dying cancer patients. Soc Sci Med 1996; 42: 721–731.[CrossRef][Web of Science][Medline]
14. Brown AD, Sandoval GA, Levinton C, Blackstien-Hirsch P. Developing an efficient model to select emergency department patient satisfaction improvement strategies. Ann Emerg Med 2005; 46: 3–10.[CrossRef][Web of Science][Medline]
15. Canadian Institute for Health Information. Hospital Report 2001: Acute Care, Ottawa: a joint initiative of the Ontario Hospital Association and the Government of Ontario, 2001. Available at http://www.hospitalreport.ca (24 May 2005, date last accessed).
16. Thompson AGH, Suñol R. Expectations as determinants of patient satisfaction: concepts, theory and evidence. Int J Qual Health Care 1995; 7: 127–141.
17. Schimert J, Schafer JL, Hesterberg T et al. Analyzing Data with Missing Values in S-Plus, Seattle, WA: Insightful Corporation 2001.
18. Fylstra D, Lasdon L, Watson J et al. Design and use of the Microsoft Excel Solver. Interfaces 1998; 28: 29–55.
19. Gesell SB, Gregory N. Identifying priority actions for improving patient satisfaction with outpatient cancer care. J Nurs Care Qual 2004; 19: 226–233.[Web of Science][Medline]
20. Gourdji I, McVey L, Loiselle C. Patients' satisfaction and importance ratings of quality in an outpatient oncology center. J Nurs Care Qual 2003; 18: 43–55.[Web of Science][Medline]
21. Liberati A, Confalonieri C, Martino G et al. Patients' assessment of quality of care: a survey of a group of breast cancer patients in Italy. Tumori 1985; 71: 491–497.[Web of Science][Medline]
22. Shilling V, Jenkins V, Fallowfield L. Factors affecting patient and clinician satisfaction with the clinical consultation: can communication skills training for clinicians improve satisfaction? Psychooncology 2003; 12: 599–611.[CrossRef][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  H. Wessels, A. de Graeff, K. Wynia, H. J. Sixma, M. de Heus, M. Schipper, G. T. G. J. Woltjer, S. C. C. M. Teunissen, and E. E. Voest Medical oncology patients' preferences with regard to health care: development of a patient-driven questionnaire Ann. Onc., October 1, 2009; 20(10): 1708 - 1713. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||