Background The objective of this study was to test the data quality, test–retest reliability and hospital-level reliability of the Patient-Reported Incident in Hospital Instrument (PRIH-I).
Methods 13 incident questions were included in a national patient-experience survey in Norway during the spring of 2011. All questions and a composite incident index were assessed by calculating missing-item rates, test–retest reliability and hospital-level reliability. A multivariate linear regression on a global item regarding incorrect treatment was used to assess the main sources of variation in patient-perceived incorrect treatment at hospitals.
Results Five of the 13 patient-incident questions had a missing-item rate of >20%. Only one item met the criterion of 0.7 for test–retest reliability (wrong or delayed diagnosis), seven items had a score of >0.5, while the remainder had a reliability score of <0.5. However, the reliability was >0.7 for six of 10 items tested at the hospital level, and >0.6 for the remaining four items. A patient-incident index based on 12 of the incident items had no missing data, the test–retest reliability was 0.6 and the hospital-level reliability was 0.85.
Conclusions The PRIH-I comprises 13 questions about patient-perceived incidents in hospitals, and can be easily and cost-effectively included in national patient-experience surveys with an acceptable increase in respondent burden. Although the missing-item rate and test–retest reliability were poor for several items, the hospital-level reliability was satisfactory for most of the items. The incident items contribute to a patient-reported incident index, with excellent data quality and hospital-level reliability.
Statistics from Altmetric.com
Patient experiences and satisfaction are important determinants of the quality of healthcare.1 Patient experiences constitute a core dimension of quality in the Organization for Economic Co-operation and Development quality indicator project, together with effectiveness and safety.2 Patient experiences and satisfaction are often measured with the aid of surveys3 ,4 that have traditionally focused on patients’ non-medical needs and preferences, such as the need for adequate information, communication and organisation.5 The safety and effectiveness components of quality have been measured using other approaches, such as mortality assessments,6 the Global Trigger Tool7 and patient-reported outcome measures.8 However, there is a growing interest in involving patients in the evaluation of their safety.9–11 Important reasons for this are an increasing focus on patient participation and empowerment in Western healthcare systems, and a broader and more complex view of safety dimensions and perspectives.
Three reviews of the literature have recently been published,9–11 two of which have a particular focus on patient reporting of safety incidents in the hospital setting.9 ,10 According to these reviews, there is currently a positive and optimistic view in the literature regarding safety reporting by patients, but also concerns about methodology and the lack of connection between patient-safety reporting systems and quality improvement. One of the reviews identified 17 relevant publications, and found that the healthcare setting, method of reporting, time span, terminology, criteria for assessment and response rate varied considerably between them.11 Another review also found considerable variation in focus, design and analysis between the 13 included studies, and concluded that from an academic perspective, patient reporting is in its infancy.9 The previous studies have not adequately documented the psychometric properties of the measurement instruments nor tested the reliability of patient-safety questions as quality indicators. Since patients struggle to understand patient-safety terminology,11 it seems especially important to test the survey questions both cognitively and using psychometric methods. Furthermore, the need to move patients’ safety reporting from the research domain to clinical governance9 includes an assessment of the usefulness and scientific adequacy of using such reports as quality indicators.
A national patient-safety campaign started in Norway on 1 January 2011. One aim for the campaign was to measure patient safety over time at different levels, including measurements of adverse events, patient-safety culture and patient reports of patient safety. Questions regarding patient-perceived incidents in hospitals were developed and tested, and these questions were included in the national patient-experience survey in 2011. The response rate for this survey has declined over time, and so the goal was to include a short instrument with little additional response burden. In addition, the new patient-incident questions to be used in the patient-safety campaign were to be included in future national reports on patient experiences in hospitals. Consequently, the incident questions were subjected to similar psychometric testing as the other experience questions. In Norway, patient-experience scales are aggregated to the hospital level and published as quality indicators in the national quality indicator system. Therefore, the new incident items also had to be tested as patient-perceived incident indicators at the hospital level.
Our standard development and validation process was the starting point for this project,12–15 including multiple activities to ensure content validity in the development phase, and a set of psychometric tests in the quantitative phase to assess instrument quality. Adjustments had to be made because of time restrictions and the measurement construct, the latter in particular resulting in substantial changes. Patient-perceived incidents are usually measured with single items, as opposed to the traditional multi-item scale approach in psychometrics. We chose to use the single-item approach because safety incidents are concrete factors, which are not well suited for the latent construct approach in multi-item scaling. However, this meant that core elements of the psychometric evaluation methods, such as factor analysis and internal consistency reliability, were inappropriate. A revised evaluation approach was thus constructed, with a focus on data quality, test–retest reliability and hospital-level reliability. Hence, the evaluation approach assessed both traditional measures of reliability and the scientific adequacy of using patient-incident questions as quality indicators. The conceptual starting point was patient-safety outcomes or incidents, as perceived by patients. The national patient-reported experience questionnaire already included one general item about incorrect treatment, and we wanted to include more detailed questions about types of incidents. In addition, broader safety dimensions like trust in doctors and information received were already part of the national patient-experience questionnaire. We aimed to include important safety issues in hospitals, but at the same time were sensitive to the fact that most patients lack medical knowledge and might find safety incident reporting difficult. The lack of a comprehensive theoretical model implied that conceptual development and refinement were expected as a result of the qualitative development activities.
The objective of this study was to document the development, psychometric testing and hospital-level reliability testing of the Patient-Reported Incident in Hospital Instrument (PRIH-I). Preliminary questions were tested in cognitive interviews with 19 patients, and ultimately 13 incident questions were included in a national patient-experience survey in Norway during the spring of 2011.
Development and cognitive testing of incident questions
The patient-incident questions were developed based on a review of the literature, interviews with internal experts and health personnel, and meetings in an internal reference group for the project. Two systematic reviews of the literature were identified10 ,11 in addition to other relevant national and international publications. Our review identified important safety domains, showed the importance of having supplemental open-ended questions, and also the importance of using lay language and cognitive testing with patients. In addition, the conceptual restriction to outcomes or incidents was challenged, since the review indicated the importance of both prevention of safety incidents and health personnel's actions after an incident has occurred.
We conducted five semistructured interviews with internal experts and health personnel with a focus on relevant topics for question generation and considerations of which topics patients are able to answer. Many topics were the same as identified in the review, and advice was also given on topics to exclude from the questionnaire. For instance, this related to psychological and socio-economical problems following an incident (the cause could both be the underlying disease and the incident), and failures with technical equipment since technical issues are difficult to evaluate for patients. Furthermore, broader safety aspects like confidence in doctors and information were stressed, but these issues were already included in the national patient-experience questionnaire. The interviews showed that prevention of incidents also should be included, in line with findings from the review.
The findings from the review and the qualitative interviews were presented for the internal reference group, which supported the expansion to prevention of incidents and questions on how health personnel handle mistakes after they have occurred. The reference group was asked to rate the three most important safety incident domains and the consensus in the group was that the following topics were the most important: (i) wrong medications or other medication-related errors; (ii) administrative errors; and (iii) hospital infections. A set of safety questions was presented for the reference group, following some revisions and a final agreement on a set of preliminary questions for further testing.
The preliminary set of questions was tested in cognitive interviews with 19 patients visiting a university hospital in Norway. The interviews demonstrated that many patients felt unable to judge certain safety aspects, which meant that some questions had to include a ‘do not know’ option. The interviews also demonstrated that some of the safety questions were irrelevant for many patients, such as medication-related errors for patients not using medicines. Many of the patients perceived that no special incidents occurred, implying that a range of questions had to be answered even though they had nothing particularly negative to report about patient safety. This meant that some questions had to include a ‘not applicable’ option.
Data collection in the national survey
The details of the national survey have been described elsewhere.16 In short, it included 400 randomly selected adult inpatients who were discharged from each hospital in Norway between 1 March and 22 May 2011. Of these, 744 patients were not eligible, so that a final cohort of 23 420 patients was included in the study. The inclusion period was divided into three 4-week groups, meaning that patients received the questionnaire approximately 1–5 weeks following discharge. We wanted experiences to be as fresh as possible, but the timing question had to take account of practical issues relating to the fact that 61 hospitals had to transfer data to the Knowledge centre for each 4-week group. The response rate was 46.4%. A retest questionnaire was also mailed to 270 consenting patients approximately 1 week after their first reply for the purpose of assessing test–retest reliability; 163 patients returned the retest questionnaire (60.4%).
The Data Inspectorate and the Norwegian Ministry of Health and Care Services approved the survey.
Questionnaire used in the national survey
Based on the development project, 13 patient-incident questions were included in the patient-experience questionnaire.17 ,18 The questionnaire used in the national survey comprised 73 closed-ended items. Most experience items had a 5-point response format ranging from 1 (‘not at all’) to 5 (‘to a very large extent’). A total of 35 items related to patient experiences with structures, processes and outcomes of healthcare were aggregated to the 10 quality indicators in the national report:19 waiting time (one item), physical hospital standard (six items), next of kin (two items), organisation (four items), doctor services (seven items), nursing services (seven items), information (three items), discharge planning (two items), cooperation with other health services (two items) and incorrect treatment (one item). The questionnaire also included an open-ended question on the last page asking for comments about their hospital stay or the questionnaire itself, and probed for information regarding errors or unnecessary problems during or after their hospital stay.
The 13 patient-incident questions and response categories are presented in online supplementary appendix 1. The topics covered were safety incidents, type of mistake, safety communication, medicine lists, infections, safety actions by health personnel such as hand hygiene and control of identity, and satisfaction with how health personnel handled the mistake after it occurred. Seven of 13 patient-safety questions had the same 5-point response format as described above, three questions had a 3-point format (‘no’, ‘once’ and ‘more than once’) and three questions had two response categories (‘yes’ and ‘no’).
An incident index consisting of 12 safety items was computed. The item about bringing a medicine list at admission to hospital was excluded because it partly relates to processes outside hospital control. For items with 5-point scales, the three worst response categories were given a value of 1, and the others were assigned a value of 0. For items with 3-point scales, the two worst response categories were given a value of 1, and the remaining was assigned a value of 0. For items with two response categories, the negative and positive response categories were given values of 1 and 0, respectively. Therefore, the incident index varied from 0 to 12, with higher numbers indicating more patient-perceived incidents.
Descriptives, missing-item rates and test–retest reliability were calculated for all patient-incident items and the incident index. The intraclass correlation coefficient was used as an estimate of the test–retest reliability for continuous variables, and the κ statistic for dichotomous variables. A widely accepted criterion is that the estimated coefficients should exceed 0.7.20
Hospital-level reliability concerns the proportion of the variance in hospital-level mean scores that is due to true differences between hospitals, as opposed to differences that might be due to sampling errors.21 Hospital-level reliability was estimated for items with 3- and 5-point response scales using one-way Analysis of variance (ANOVA) to distinguish the between-hospital variance from the within-hospital variance. The intraclass correlation coefficient is calculated as the variation between hospitals divided by the total variation. The estimated reliability at the hospital level in the present study can be used to estimate how large samples in other studies would need to achieve specified levels of hospital reliability, such as how many respondents are needed to reach the standard reliability criterion of 0.7. These estimates were calculated using the Spearman–Brown prophecy formula.
Multivariate linear regression analysis was used to assess sources of variation in a global item about incorrect treatment, and included all specific safety items as predictors. The qualitative development work identified the importance of other safety-related experiences for patients’ feeling of safety in hospitals, and these were also included as predictors in the regression: information (scale, three items), doctor services (scale, seven items), nursing services (scale, seven items) and organisation (three items). To avoid extensive loss of cases and generalisability, items with a large amount of missing data were recoded and included in the regression as dummy variables. For example, the unnecessary damage related to surgery was recoded into two dummy variables: (i) patients answering ‘not at all’ or ‘to a small extent’ and (ii) patients not responding to the item. The reference group was patients answering ‘to some extent’, ‘to a large extent’ or ‘to a very large extent’, and regression coefficients should consequently be interpreted as the partial effect of being in the respective group compared with the reference group.
All analyses were conducted using SPSS (V.15.0), except for reliability analysis at the hospital level, for which Microsoft Office Excel 2007 was used.
Of the cohort of hospital patients, 10.5% perceived that they had been incorrectly treated by the hospital (table 1) ‘to some extent’ (6.4%), ‘to a large extent’ (2.3%) or ‘to a very large extent’ (1.8%). Most items had a large majority of responses in the most positive response category, indicating a high level of patient-perceived safety in the hospital. The most negative responses were found for questions about control of medicine lists, satisfaction with how health personnel handled the mistake or problem when it occurred, and transfer of important information between health personnel. The mean score for the patient-incident index was 1.21 (SD 1.69).
The missing-item rate exceeded 20% for five of the 13 patient-incident questions (table 2). The item about satisfaction with how health personnel handled the mistake or problem was unanswered by 70.9% of patients, while item-missing was 43.6% for the question about bringing an updated medicine list at admission. The missing-item rates were the lowest for questions regarding control of identity (2.7%), administrative mistakes (3.0%) and receiving important information (3.1%). There were no missing cases for the incident index.
Only one item met the criterion of 0.7 for test–retest reliability, namely, that about wrong or delayed diagnosis (table 2). Seven items had a test–retest reliability greater than 0.5, ranging from 0.67 for the item about updated medicine list check at discharge to 0.51 for the item about control of identity. The remaining items had a test–retest reliability rating of less than 0.5, ranging from 0.44 for hospital infections to 0.16 for the item about receiving important information from hospital staff. The test–retest reliability rating for the incident index was 0.61.
The reliability rating exceeded 0.7 for six out of 10 continuous items at the hospital level, ranging from 0.85 for the item about transfer of important information between health personnel to 0.70 for the item about unnecessary injury/problem related to surgery. The remaining four continuous items had a hospital-level reliability of between 0.6 and 0.7, while the hospital reliability was 0.85 for the incident index. With the exception of the item about control of identity, fewer than 200 respondents were needed for each hospital to reach the reliability criterion of 0.7. A reliability criterion of 0.9 would require very large hospital samples for all incident questions (range 256–1014), but the incident index and two items reach 0.9 with less than 300 respondents. Figure 1 shows the mean score for the patient-incident index by hospital, which varied from 0.54 for the best hospital to 1.68 for the worst.
Multivariate linear regression showed that unnecessary injury following surgery (p<0.001), wrong or delayed diagnosis (p<0.001) and medication-related errors (p<0.001) were the most important predictors for perceived incorrect treatment (table 3). In all, 12 out of 17 predictors were significantly associated with perceived incorrect treatment, which also included experiences related to doctor services, organisation and information. The regression model explained 39% of the variance.
There is a growing interest in involving patients in the evaluation of their safety.9–11 However, previous studies have not adequately documented the psychometric properties of the measurement instruments nor tested the reliability of patient-safety questions as quality indicators. The goal of this study was to document the development, psychometric testing and hospital-level reliability testing of the PRIH-I. The PRIH-I consists of 13 questions about patient-perceived incidents in hospitals and was tested in a national patient-experience survey. The missing-item rate and test–retest reliability for several items were poor, but hospital-level reliability was satisfactory for most items. The incident items contribute to a patient-reported incident index, with excellent data quality and hospital-level reliability.
The national patient-experience survey about hospitals was conducted among random samples of adult inpatients from all hospitals in Norway. This generic approach to patient-perceived incidents includes patients across all hospital departments for adults. We are only aware of one previous study that has used such a generic approach, but that study was limited to a single hospital.22 The present study was conducted among all hospitals in Norway, and the standardised methods ensure comparability between hospitals. This means that patient-perceived incidents can be aggregated and analysed at both the national and hospital levels, together with other quality indicators. The patient-incident index appears to be especially promising in a quality indicator setting, since it scored highly on all tests applied in this study.
However, both the development activities and empirical testing demonstrated that many hospital patients find patient-safety questions difficult to answer and/or of little relevance. For example, items about surgery are not relevant for patients who do not receive surgical interventions. This means that many patients have to respond to irrelevant questions by ticking ‘not applicable’ or the top box response option. The national survey showed that the amount of item-missing was large for some items, but this is a natural reflexion of relevance across the patient population. Combined with the declining response rate in the national survey, this meant that the goal was to develop and test a short incident instrument. The PRIH-I comprises 13 items, and the response rate in 2011 was similar to that of the last survey in 2006. This indicates that the inclusion of the PRIH-I in the national survey was acceptable to patients.
The importance of using lay language for patients reporting about their safety has been stressed previously.11 We used cognitive interviews to test incident concepts and questions with patients. In psychometrics, items are considered to be empirical reflections of a latent construct, and multi-item scales are the standard approach. The reliability is normally lower for single items than for multi-item scales, which was also the case in the present study. All patient-experience scales had reliability estimates exceeding 0.7,19 but only one of 13 incident items met this criterion. The low level of test–retest reliability for many items shows the presence of substantial amounts of measurement error. This might be related to cognitive difficulties that remain with the safety questions, a lack of stability in the construct over time because of actual changes23 or changes in patients’ internal values.24 For instance, the question about staff forgetting to give important information to the patient only had a test–retest reliability on 0.16, implying that more than 80% of the variation between individuals is related to measurement error. The question is more general than the other questions, and the lack of specificity might have resulted in some patients evaluating different experiences in the test and retest questionnaires. More generally, it is important to note that the test–retest estimates are based on a rather small sample. In addition, the amount of variation on some variables is small, implying that relatively large effects might be caused by changes within a few individuals. It is not possible to determine the exact causes for individual changes across the set of dependent variables. However, the practical implications are to focus on the safety index (and possibly the most reliable single items) and to secure an adequate sample size at the hospital level. The patient-incident index consisting of 12 incident items had acceptable test–retest reliability, and hence constitutes a robust alternative to single items. Furthermore, when there were sufficiently large samples per hospital (n=200), all except one item reach the criterion of 0.7 for hospital-level reliability. The choice of outcome variables and sample size are important considerations when planning and designing a study of patient-perceived incidents. Based on this study, it seems advisable to recommend around 200 respondents per hospital and to use the incident index as a primary measure when comparing hospitals and monitoring changes over time.
The findings of this study demonstrate that patients have a broad perspective about hospital safety. The multivariate linear regression found that unnecessary injury following surgery, wrong or delayed diagnosis, and medication-related errors were the most important predictors for perceived incorrect treatment. However, organisation of hospitals, information and doctor services were strongly related to patient perception of overall incorrect treatment. This means that a full understanding of the patient perspective about hospital safety should include such dimensions. It also indicates that safety improvement work should be directed towards these more general quality issues. The percentage of patients perceiving incorrect treatment in Norwegian hospitals is stable over time. The percentage is somewhat lower than the estimate of 16% of hospital stays in 2011 with at least one adverse event advent, estimated by the national safety campaign in Norway using the Global Trigger Tool (GTT) methodology. However, the quality of the GTT measurement in Norway has not been evaluated formally. The change over time in patient-based estimates and convergence with other sources is an important future research area.
Several previous studies have compared patient-safety incident reporting against other sources.25–29 While this is a generally relevant research area, it has little relevance for quality improvement efforts. Patient-reported safety and clinical patient safety are different concepts;10 the aim of the former is to improve our understanding and measure safety from the patient perspective, which is best evaluated by the patients themselves, while that of the latter is to assess safety in medical terms, which is best evaluated by medical professionals. Patient-reported incidents that are not confirmed by other sources are still quality problems: lack of convergence indicates communication problems with negative emotional consequences for patients, and hence poor patient-centeredness. Patient-centeredness is an important part of healthcare quality.1 ,2 Furthermore, it is not clear what to use as a gold standard for adverse events. For example, recent research on the inter-rater reliability of the Global Trigger Tool has shown that assessments by even experienced healthcare teams can differ substantially when there are adverse events in the same sample of records.7 All in all, the optimal solution seems to be to triangulate different methods and perspectives, and have a broad and complex view of safety measurement and improvement.
The timing of patient reports is considered an interesting research area.9 Many studies of patient experiences have found that making measurements at the time or shortly after discharge underestimates the amount of problems, because of social desirability bias and other sociopsychological factors.30 Timing research needs to assess the appropriate balance between multiple factors: the need for hot reporting and safety improvement, sociopsychological pressures for social desirability bias, patient recovery and adequate mental distance to the incident, and the potential for recall bias. Patient-incident reporting at home after hospital discharge obviously increases the likelihood of underreporting of the most serious mistakes and deaths. To compensate for this, the patients’ next-of-kin and other proxies should be able to respond on the patients’ behalf. In addition, other safety reporting systems must be used to ensure full coverage of all safety incidents.
The lack of a comprehensive theoretical model is a limitation of this study. The conceptual approach was further developed and refined as part of the project, but the study did not follow our standard development and validation process.12–15 An internal reference group was used instead of an external group, and we only conducted cognitive interviews with patients, not indepth interviews. The inclusion of the PRIH-I within an existing patient-experience questionnaire gave rise to space restrictions. The existing patient-experience questionnaire was not developed and validated with a particular focus on patient-perceived safety, which means that broader dimensions of patient-perceived safety might be poorly represented. Furthermore, topics like falls, pressure sores and equipment failures are not included in the PRIH-I. Some of these incidents were excluded because they were difficult for patients to evaluate, but it nevertheless means that the PRIH-I does not have full coverage of all possible incident types. Future research could test other ways of formulating such questions to improve content validity.
The psychometric evaluation model differed from the standard approach by focusing on single items instead of multi-item scales. A single-item approach means that factor analysis and internal consistency reliability are not relevant. The revised assessment method included test–retest reliability and hospital-level reliability for all safety items, as well as the construction and testing of an incident index consisting of 12 incident items. We believe this to be a robust model, but stress the importance of using the Index and an appropriate sample size when the PRIH-I is applied at the hospital level. Furthermore, more research is needed on the PRIH-I, including its ability to measure changes over time, correlation with other quality and safety indicators at the hospital level, case-mix considerations, and research into how the PRIH-I can be used effectively in local quality improvement work. The last topic is especially important since one of the aims of national patient-experience surveys in Norway is to improve quality.
The PRIH-I consist of 13 questions about patient-perceived incidents in hospitals, and can be easily and cost-effectively included in a national patient-experience survey with only a small increase in respondent burden. Although the missing-item rate and test–retest reliability for several items were poor, hospital-level reliability was satisfactory for most items. The incident items contribute to a patient-reported incident index, with excellent data quality and hospital-level reliability.
Thanks to Kari Aanjesen Dahle for being project leader for the development project. Thanks to Marit Skarpaas, Sinan Akbas, Ulla Benedicte Funder and Solveig Eggen for their contribution to the national data collection, and Tomislav Dimoski for developing the FS system, carrying out the technical aspects of the national survey and being project leader for data collection.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Data supplement 1 - Online appendix
Contributors OB planned the paper together with KES, HHI and AKL, carried out the statistical analysis, and drafted the paper. KES, HHI and AKL participated in the planning process, revised the draft critically and approved the final version.
Funding The study was financed by the Norwegian Knowledge Centre for the Health Services.
Competing interests None.
Ethics approval The Data Inspectorate and the Norwegian Ministry of Health and Care Services approved the survey.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement Anonymous data can be made available upon request.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.