KARDEMİR KARABÜK DEMİR ÇELİK

Assessment of generalisability, or external validity, is important in any study to help clinicians decide whether the findings are applicable to a wider range of patients[187,202].

Lack of consideration of external validity is a frequent criticism of trials, reviews, and guidelines[187]. Although generalisability has been highlighted in the various papers

constituting this PhD, its importance warrants a more concise debate about the applicability of the results.

The setting of the trial includes the health care system, country, and selection of participating centres and clinicians (see Chapter 2.6). The current results are likely relevant for patient care pathways in Norway, but also in other health care settings in which GPs have

65

a gatekeeper role. In the selection of participating GP surgeries, all potential participants were included, which should increase the generalisability of the final results. The hospital was a local secondary hospital, university affiliated, but not a tertiary referral centre. It is therefore likely that similar interventions at facilities with a different case-mix may lead to different results.

The selection of patients is an important determinant of external validity (see Chapter 4.2.2). The current study utilised few exclusion criteria and included a variety of patients from ordinary clinical practice. However, as discussed in Papers I and III, we have no

conclusive indication of the percentage of the eligible patients that actually participated. The clear intent of the study to include all patients would ensure generalisability, at least within similar health care systems, but the unknown inclusion rate hampers the analysis to some extent.

Characteristics of the randomised patients also contribute to external validity. The baseline characteristics of patients in the current study did not differ significantly. However, this does not necessarily mean that they are representative of the general population referred to this, or any other hospital. The underlying pathology, severity of disease, and comorbidities may differ between the study population and the general population referred to a hospital for care within the four diagnostic groups[187]. The project scored patient care pathways, but did not register specific final diagnoses in detail. We did, however, record presence or absence of a diagnosis of incident cancer. In the group referred for suspected colorectal cancer, the overall cancer rate was 8.4%, and in the dyspepsia group it was 0.8%.

This corresponds well with findings from colonoscopies in a screening programme in the United Kingdom[209] and with findings at gastroscopies of dyspeptic patients[80]. Of course, the population in a screening programme will not be entirely the same as ours, but those who underwent colonoscopy in the screening programme already had a positive faecal occult blood test, and hence were not that dissimilar from patients with suspected colorectal malignancy in the current study. There is no clear indication that our study population deviates significantly from the general population under care for similar conditions elsewhere in Norway or Western Europe.

66

Often in clinical trials there are differences between the trial protocol and routine practice, which may hamper the generalisability of results[187]. The current cluster randomised trial was designed to be as close to ordinary clinical practice as possible.

Although a Hawthorn effect may have affected the results, a direct transfer of the methods applied would be possible within any health care system that utilises referrals sent from a gatekeeping GP to a hospital.

In relation to outcome measures, the current trial employed novel quality criteria, together with patient experience and health care process outcomes. It is evident that the variation in quality assessment highlights some of the difficulties in assessing quality of care at a practical, patient-centred level. This variation may very well be seen as a limitation to the generalisability of the outcome measures. However, the options do not provide an easy alternative (Paper III). Many common quality metrics are developed to measure specific actions at the hospital/GP surgery level, and do not necessarily highlight the care pathway of the individual patient[210], as envisaged in this project. The outcome measures related to prioritisation, subjective quality score, and positive predictive value of referral are likely more generalisable.

Overall, the current study design and implementation means the generalisability of the results is quite high, at least to other areas where access to specialist services is via a gatekeeping GP. The transfer of the results to other health care settings may be hampered, as the baseline quality of referral and hospital/GP communication in a small Northern Norwegian hospital may be higher than that in large health care systems.

67