To the best of our knowledge, studies about continuous cost generation of cancer prevention programmes have not been previously reported despite that many studies have acknowledged the limitations of cross-sectional cost estimates. Our study attempts to fill this gap. The gradual and continuous decline of cost indices in our study strongly indicated the ever-improving cost efficiency of endoscopic surveillance for GC through the GCEP during the observation period. This study as a free standing cost analysis, despite its inability to compute the cost-effectiveness ratio, provided important empirical evidence for programme management and ultimately for value-for-money decision making. Cost studies of long-term GC surveillance have long been anticipated worldwide given that research already illustrated the benefit of GC surveillance [27, 28]. Exploring the mechanisms underlying our results would be of universal interest to GC researchers.
Economy of scale is considered the main reason for the cost reduction, especially for the Overall Cost and the GCEP Cost [16, 17]. Previous studies have shown that as the screening volume increased, the average cost borne by the health care provider to serve one subject decreased, thereby approximating an inverse relationship . The GCEP has experienced a 7-fold increase of patient volume from 175 in 2004 to 1223 in 2009. Consequently the average cost decreased because fixed costs spread out horizontally across the number of subjects served every year and vertically along the implementation time. Correlation analysis showed a negative correlation coefficient between workload and Overall Cost and GCEP Cost, with the former achieving statistical significance (r = −0.821, p = 0.023).
More efficient utilization of the resources within the GCEP system was another factor driving the GCEP Cost down. It is well known that public health programmes can improve operational efficiency through self-learning , which would in return lead to decreased costs borne by the service provider. Having been in operation for seven years and with quality assurance protocols in place, the GCEP could optimize work-flow processes by shortening waiting times, avoiding repetitions and enhancing service awareness in team members . Although specific parameters were not set to gauge its work-flow processes, it was fair to assume that the self-learning mechanism took effect, especially in the inception of the Full Implementation phase which was the costing period of our study.
The decline in Clinical Cost indicated that subjects consumed less of the clinical services, which consist of follow-up OGD, specialist consultations, diagnostic tests and medications (Table 1). This is most likely due to a reduced demand of clinical services in later follow-ups when the subjects were experiencing fewer symptoms as a result of surveillance and associated treatment. Our findings were consistent with a cost study of a colorectal cancer screening programme which showed that repeated screenings cost less than initial screenings because of the lower prevalence of disease in the rescreening group as opposed to first-time participants . Congruent with the previous observation, the Personal Cost also decreased as patients paid less for clinical services.
A further reason for the Personal Cost reduction was the declining price of patient time estimated by the human capital approach , whereby the opportunity cost of taking one day off work for an OGD or clinic visit was measured as a single day’s salary. In Singapore, there was a large decrease in salary from the age group 50–59 years to the age group 60 years or above . As one of the GCEP inclusion criteria was being age 50 years old or above, we noted that during the observation period, 42 subjects (19.4%) underwent the age change from 50–59 years old to 60 years old and above.
Subgroup analysis was conducted to explore the cost generation in subgroups categorized by age and gender, which are relevant to the diagnostic yield of a screening programme [34, 35]. Similar to the observation for the whole sample, both gender subgroups experienced significant annual decreases with bigger decrements in females for all four cost indices (Table 3). As for the impact of age, compared with subjects 60 years or older, subjects between 50 and 59 years had a larger decrement in Overall Cost and Clinical Cost, US$109 vs. US$102 and US$41 vs. US$37 respectively, and a smaller decrement in the GCEP Cost and Personal Cost, US$40 vs. US$59 and US$7 vs. US$10 respectively (Table 3). Comparing the average costs of the subgroups for either variable failed to reveal significant differences, as illustrated by the overlapping curves in Figures 2 and 3. The cost efficiency in subgroups as described above was of great significance in advising resource distribution among these subgroups and in computing population specific cost-effectiveness ratios subsequently.
As the cost-effectiveness of screening is sensitive to disease incidence in target populations [11, 36], the GCEP classified subjects into high and moderate risk of GC, which subsequently determined the frequency of surveillance OGD. The temporal trends of cost indices were statistically different between high and moderate risk groups (Figure 4 and Table 3). The Overall Cost and the Clinical Cost for the high risk group had annual decrements 2.4 times lower than those for the moderate risk group. The GCEP Cost and Personal Cost showed a downward trend in the high risk group, while they both increased over time in the moderate group. The cost difference between high and moderate risk groups was arbitrary as OGD frequencies were decided beforehand, yet it has implications for funding and for evaluating the cost-effectiveness of a specific population.
Compared to other published cost-analyses of cancer screening programmes, our study was unique in four ways, in that we 1) analyzed long-term continuous cost generation; 2) collected individual-level data; 3) identified and quantified all possible resources; and 4) studied multiple indices simultaneously. The advantages of these are discussed as follows.
Given that long-term or life-long follow-up is required in a cancer surveillance programme we studied a prospective cohort, the GCEP, with 6.5-year follow-up data and reported on the temporal trends of cost indices, in addition to the point estimates which are the sole outcomes in cross-sectional studies [13, 17]. There is a high likelihood that point estimates are skewed or biased depending on the period chosen in a specific study [14, 37]. Programme activities and patient volume varied greatly from year to year resulting in inflated/deflated point estimates [13, 38]. Our study, rather than overestimating/underestimating the cost values, reported on the temporal variation of costs that can be used to predict the variability and the evolution of the cost - two aspects crucial for programme budgeting.
Regarding the quality of data, our study collected individual-level data based on the NUH GCEP database. The quality of our data afforded statistical advantages over aggregate data analyzed in other studies [13, 39]. Individual-level data captured person-to-person and year-to-year variations which allowed us to estimate the means and confidence intervals from actual distributions and to apply GEE models, thereby enhancing the validity and reliability of our results.
In addition, we used the original documents of patient casenotes and the GCEP financial statements to identify clinical and non-clinical items directly associated with programme operation. A bottom-up approach was adopted to quantify resources consumed and to estimate their monetary value [22, 40], thereby avoiding subjectivity or recall bias when data is collected through a survey-based top-down approach , and ensuing high accuracy and completeness of the data.
A major contribution of our study was that we simultaneously investigated multiple cost indices, each of which has been a focus in separate previous studies [13, 16, 32, 41]. To our knowledge, no study has investigated these indices simultaneously in a single study thereby overlooking the fact that these costs accrued concurrently. Complete and accurate cost data are crucial to both an economic evaluation and programme planning. Economic evaluations tend to underestimate the cost because of poor representation of personal costs and programme costs . The personal cost represents the financial commitment of a subject to participate in screening [20, 41], so it was associated with subject compliance and programme effectiveness [43, 44]. Our study found that patients paid at the most 18.2% of what was borne by the service provider (the GCEP) (Figure 1), suggesting that the co-payment could be a viable arrangement. Programme cost measures the expenditure on non-clinical activities and represent the internal resource allocation within programmes. A cost analysis of a colon cancer screening programme demonstrated that non-clinical activities consumed more than 50% of the total budget , exceeding the US federal standard of 40% . In our study, the Clinical Cost accounted for only 17.35% to 35.76% of the Overall Cost, i.e., the non-clinical cost ranged between 64.24% and 82.65% (Figure 1). Although this study took a societal perspective and applied a narrower definition of clinical service, as an organized surveillance programme in a small country such as Singapore, a high proportion of non-clinical expenditure appealed to the more efficient internal resource allocation.
We acknowledge several limitations with our study. As a pure cost analysis, this study is inherently unable to inform the value-for-money decision which is of utmost importance yet requires a full economic evaluation. In addition, service underutilization which is negatively associated with programme effectiveness, cannot be ruled out as a mechanism driving down the cost in our study. The co-payment system whereby patients are committed to a certain amount of money could impede some patients from using GCEP services, especially those from low-income families [18, 44, 46]. Removal of patient costs has been demonstrated to increase the screening compliance . Retrospective data collection in the current study cannot accurately match the cost with the specific clinical or administrative activities. Therefore, we could not identify the area of inefficiency. As for the Personal Cost, we may have omitted some elements which could only be retrieved through personal interview. Furthermore, caution is needed to extrapolate the downward trends beyond the observation period, because all the factors accounting for the cost reduction have limits . Nonetheless, our results confirmed continuous cost decrements in the early phase after full implementation. However, data seemed to indicate that the descending momentum has stopped in 2009 (Figures 1, 2, 3 and 4). The ideal situation is that a programme achieves its optimal cost efficiency and functions on its minimum average cost curve . A measure of a successful programme is how soon this point is reached, however this was not captured in our study.