We developed an online survey instrument (see online supplementary appendix A1) modelled after the 2003 survey.15 With respect to costs, the instrument requests annual estimates for 11 major cost categories (see below) each for analysis of samples obtained through AFP surveillance and from ES. For the cost categories ‘equipment’ and ‘durable supplies’, we asked for annual amortised costs, defined as purchase, packing, freight and insurance costs divided by the expected useful lifetime, and we provided a spreadsheet to help respondents compute the annual amortised costs. In addition, for laboratories that recently (ie, between 2010 and 2016) established or significantly expanded their ES capacity, we requested estimates of the ES set-up costs for 10 largely overlapping cost categories relevant to establishing ES capacity. For all of these, we asked respondents to provide the breakdown of costs by funding source (ie, internal, external (GPEI), bilateral (non-GPEI, non-national)). The instrument further included questions about the role and capacities of the laboratories, geographical areas served, staff time spent on different activities, number of samples processed for different tests (eg, virus isolation, ITD, sequencing and, for ES samples, concentration), serological testing activities, non-polio surveillance activities supported by funding for polio (ie, polio-supported staff), the nature of ES activities and anticipated future changes in workload or workflow.

We piloted the survey among all WHO regional coordinators of the GPLN and a small subset of laboratories before launching the revised, final instrument online and in PDF form in July 2017, in English, Chinese and Russian. We targeted all 145 active GPLN laboratories (we excluded one laboratory considered dormant) and three non-GPLN laboratories recently established to facilitate ES in countries with no easy access to a GPLN laboratory for sewage sample concentration and processing (ie, concentration-only laboratories). We followed up with responding laboratories to resolve any ambiguities or apparent inconsistencies in the responses (see online supplementary appendix A2 for a list of the responding laboratories). We reached out four times to non-responding laboratories to increase the response rate through November 2017 and closed the online survey instrument at the end of 2017.

We collected all original responses directly from the online survey instrument and manually entered any changes indicated by respondents during the follow-up. For rare instances in which a laboratory provided a range of costs for a category, we used the midpoint. Some respondents noted that they reported costs for consumable supplies or shared consumable supplies on a per-sample basis rather than as an annual total, which prompted us to systematically convert consumable supply costs to annual totals when we suspected responses per sample (see online supplementary appendix A3). We converted all monetary estimates to 2016 US$ using publicly available exchange rates from 1 July 2016.18 We classified the income levels of laboratories based on the 2016 World Bank income levels of their host countries.19 Unless otherwise noted, all results represent the annual totals for 2016.

To account for missing cost responses from responding laboratories, we interpreted unanswered or zero responses differently depending on the cost category. We assumed that all laboratories incur costs under the six cost categories of personnel, equipment, durable supplies, consumable supplies, operations and shipping/transport (ie, non-zero categories). In contrast, we assume that some laboratories may truly not incur any costs for the five categories of training, shared consumable supplies, donated supplies, technical support and other (ie, possible zero categories). Furthermore, we preprocessed some of the cost data before further analysis because some respondents indicated challenges in separating costs between analysis of AFP and ES samples and others explicitly indicated that they reported only the combined costs. Compared with samples from patients with AFP, the processing of ES samples follows a more involved algorithm (ie, three times as many cell cultures),16 more often yields viruses that require ITD testing or sequencing (ie, because an ES sample represents a composite sample from many individuals) and requires about four times the processing time by trained staff.20 The type and nature of adjustment depended on the nature of the missing data (see online supplementary appendix A3).

To account for non-responding laboratories, we considered variables that we could obtain outside of the survey for all laboratories from the web-based GPLN management system, including number of employee full-time equivalents (FTE) employed for poliovirus surveillance, and number of virus isolation tests, ITD tests and sequences performed on AFP samples. Based on differences between laboratories and descriptive analysis of relationships by WHO region, income level and laboratory role, we grouped the laboratories by income level and capacity (ie, virus isolation only, ITD and virus isolation but no sequencing, and sequencing (with or without ITD capacity)) for regression analyses. Within each group, we used univariate linear regression on the number of samples processed for virus isolation to estimate missing costs. In the event of negative intercepts or slopes in a given cost category and group, we forced the intercept to 0, thus effectively reverting to estimation based on the simple average cost per sample processed for virus isolation for the given cost category and group. We also considered linear regression on the number of FTEs, multilinear regression on all variables and different grouping approaches, but found no substantial improvement or differences in the totals.

To estimate the costs of analysis of serum samples, we assume costs of $10 per sample for consumables and equipment. For the personnel costs, we multiply the reported average personnel costs per FTE in upper middle/high-income countries (since these countries test most of the reported serum samples) by the reported number of FTEs for processing of serum samples. We estimate the costs of research and development activities based on extrapolation of data from the largest global specialised laboratory (ie, the US Centers for Disease Control and Prevention (CDC) laboratory in Atlanta, GA) (MAP, MSO). We estimate the global overhead costs for coordination, training and technical support not incurred by individual laboratories based on WHO surveillance budgets (OMD).

This survey did not involve patients or public opportunities for engagement.

We received responses from 132 of 149 (89%) surveyed laboratories, which included one concentration-only laboratory. Figure 1 provides the breakdown of the response rate by laboratory role, region and income level, which shows a response rate of at least 78% for all breakdowns, except for the three concentration-only laboratories, from which we received only one response (ie, response rate 33%). Based on the reported capacities, we grouped the 131 responding GPLN laboratories into 30 (23%) laboratories with virus isolation capacity only, 67 (50%) laboratories with virus isolation and ITD capacity and 35 (27%) laboratories with sequencing capacity (regardless of virus isolation and ITD capacity), with the concentration-only laboratory equipped with neither of those capacities. For the estimation of costs to process AFP samples, we further grouped the laboratories by income level into low/lower middle income versus upper middle/high income to allow more appropriate cost extrapolation while maintaining sufficient numbers of laboratories in each group. For the estimation of costs to process ES samples, we did not stratify by income level because of the smaller numbers of laboratories in this group.

Table 1 summarises the breakdown in the laboratory types and the numbers of laboratories in each category. The reported costs to process samples from AFP cases and contacts for each individual cost category reflect different response rates for the various categories (table 1, numbers in parenthesis next to the reported costs show the per cent of laboratories reporting). The reported costs for each category remained markedly lower than the overall survey response rates (compare figure 1 with table 1), and show the highest reporting percentages for personnel and consumable supplies. The responding laboratories reported approximately $16 million in total AFP-related costs (table 1), which does not include $510 000 in reported AFP-related costs from 12 laboratories that we reallocated to processing of ES samples. Personnel accounted for 44% of all reported costs, followed by consumable supplies (21%) and equipment (20%).

Figure 2A shows the source of funding by cost category for the costs reported. Internal (national) funds accounted for a large proportion of personnel (76%), training (66%), equipment (64%), operations (79%) and technical support (85%) costs, while external (GPEI) funds accounted for a large proportion of costs for consumable supplies (72%), donated supplies (75%) and shared consumable supplies (54%). Overall, 61%, 36%, 2.4% and 1.3% of all reported funds to process AFP samples came from internal, external, bilateral and unspecified funds, respectively. Twenty-six per cent of laboratories reported dependence on non-internal funds for at least 50% of their budget, with regional percentages of 0%, 3.3%, 6.7%, 50%, 58% and 86% for the American, Western Pacific, European, Eastern Mediterranean, Southeast Asian and African WHO regions, respectively.

Finally, table 1 (bottom section) also reports the total costs estimated for each laboratory group and cost category, based on extrapolation to the entire network of laboratories. The resulting total AFP costs equal approximately $28 million. Although the sequencing laboratories account for only 26% of the total number of GPLN laboratories, they account for 34% of the estimated lab-specific costs for processing of AFP samples.

Fifty-nine (45%) of all 132 responding laboratories reported supporting ES activities, including one concentration-only laboratory. One additional laboratory that reported not analysing ES samples estimated the costs of supporting national ES activities with a staff member providing technical support. We excluded the latter laboratory and the concentration-only laboratory due to the absence of numbers of ES samples processed for virus isolation needed for inclusion in the regression. Seven non-responding GPLN laboratories support ES according to unpublished WHO data, leading to a total of 65 (45%) GPLN laboratories supporting ES activities in 2016.

Table 2 shows the reported and estimated recurring costs for ES based on the variable response rates for each cost category. The responding laboratories reported approximately $3.2 million in total recurring ES-related costs, which includes $510 000 in AFP costs that we attributed to ES. Varying the ratio of ES processing cost per sample to the AFP processing cost per sample from 3 to 10 changed the AFP processing costs attributed to ES processing from $340 000 to $590 000, respectively. Thus, the impact of this assumption on overall costs remains modest, because it only affects 12 laboratories with ambiguity about whether reported AFP processing costs included ES processing costs. The breakdown by cost category remained similar to the costs for processing of AFP samples, and similarly the sequencing laboratories accounted for a large portion (58%) of all reported recurring ES costs.

Figure 2B shows the breakdown by cost category and funding source for the reported costs in table 2, which shows a similar breakdown as for AFP sample processing costs. Overall, 65%, 22%, 0.3% and 12% of all reported recurring ES costs came from internal, external, bilateral and unspecified funds, respectively.

The bottom half of table 2 shows the extrapolated costs estimated in each group and for each cost category. The resulting total recurring ES costs equal approximately $5.3 million. Table 2 does not factor in the relatively small costs from the one concentration-only laboratory that responded to the survey, which reported only some internally funded recurring ES costs for personnel with other costs captured in the ES set-up costs or unquantified because they paid for by external resources.

Of the 59 laboratories (ie, 58 GPLN laboratories and one concentration-only laboratory) that reported supporting ES activities, 35 (59%) reported that they recently (ie, between 2010 and 2016) set up or significantly expanded their ES capacity. Of these 35 laboratories, 25 (71%) provided set-up cost estimates for at least one cost category, leading to total reported set-up costs of approximately $1.8 million. This includes estimates from 16 ITD laboratories, 6 sequencing laboratories, 2 virus isolation laboratories and 1 concentration-only laboratory. Only 6 of the 25 (24%) laboratories reported becoming fully operational during 2016, which suggests that most of the reported set-up costs did occur sometime between 2010 and 2015. Figure 3 shows the breakdown of the $1.8 million of reported ES set-up costs, with the legend also showing the response rates for each set-up cost category. New equipment for concentration represented the largest contributor to all reported set-up cost (38%), followed by new equipment for expanded poliovirus processing capacity (12%), new personnel (12%), new consumable supplies (11%) and facility costs (10%). These results suggest that establishing new ES capacity in a laboratory costs approximately $75 000.

We explored the breakdown of reported staff time spent on polio and non-polio diseases by WHO region for staff supported by funding for polio (see online supplementary appendix A4). We also characterised the reported number of samples or isolates processed in the context of different activities (see online supplementary appendix A4), with the approximately 250 000 samples from AFP cases and their contacts processed for virus isolation dominating the results and reflecting the primary focus of the GPLN on supporting AFP surveillance. Given the current prevalence of wild polioviruses and level of OPV use, roughly 4.5% of stool samples from AFP cases grow in the L20 B cells used for virus isolation. Of these, approximately 7% appear as possible wild or vaccine-derived poliovirus, which then undergo sequencing. In contrast, ES accounted for only 12 000 samples processed for virus isolation originating from 8200 environmental sample concentrates, 67% of which were concentrated using the WHO-recommended two-phase method.16 The difference between the number of concentrates and the number of isolates probably comes from laboratories that (re)tested samples already concentrated by another laboratory, including third-party laboratories not part of the GPLN.

Table 3 estimates the full polio laboratory costs for 2016 based on the results of the survey complemented with data from the WHO and the CDC global specialised laboratory in Atlanta, GA. Using the results from tables 1 and 2, we estimate the total laboratory-specific costs to support AFP surveillance and ES at approximately $33 million. This does not include the reported recent ES set-up costs of $1.8 million, which represents only a fraction of the WHO-supported ES set-up costs for 2016, or the costs for the analysis of serum samples. For 2016, we estimate total costs of serology of approximately $1 million, total costs of research and development activities of approximately $3 million and global overhead costs for coordination, training, technical of approximately $6 million. The resulting estimated total poliovirus laboratory costs for 2016 equal to $43.3 million.