Study design and ethical approval
This was a retrospective, laboratory-based study performed at the Division of Medical Virology, Department of Pathology, National Health Laboratory Service (NHLS) Tygerberg and Stellenbosch University. Ethical approval was obtained from the Health Research Ethics Committee of Stellenbosch University with reference number N14/02/013.
In order to assess specimen flow and positivity rate in the abovementioned laboratory, we queried routine PCR data from the Disa*Lab laboratory information system (Laboratory System Technologies (Pty) Ltd., Bedfordview, South Africa). Routine HIV PCR data was obtained by extracting the following parameters from the local Medical Virology NHLS database: unique laboratory reference number, patient name and surname, patient date of birth, sample processing date, sample result. Patient records were linked using LinkPlus software (CDC, Atlanta) to determine the final HIV status of presumed low positive samples, where after all records were de-identified.
Reference samples used for pooling evaluation
Per laboratory standard operating procedure, the laboratory routinely stores residual EID samples as 50 μl DBS after routine diagnostic testing using 100 μl EDTA whole blood on the Roche Cobas AmpliPrep/Cobas Taqman HIV-1 Qual (CAP/CTM, Roche Molecular Systems, Inc., Branchburg, NJ) assay has been completed. The Roche CAP/CTM is a total nucleic acid real-time PCR assay that detects HIV-1 proviral DNA and HIV-1 RNA . The assay is suited for EID testing in high burden areas as it is a high-throughput, automated system, which can test both whole EDTA blood and DBS samples.
Stored DBS are identified by unique laboratory reference numbers. Previously confirmed negative DBS, tested in single reactions with the reference PCR method, were combined to constitute negative pools. Positive pools were constituted by combining one DBS from a patient with a positive reference result with negative dried blood spots. A subset of the positive pools were based on ‘low positive’ samples, i.e. samples initially categorised as indeterminate (CAP/CTM cycle threshold values ≥32 and/or relative fluorescence values ≤5) but who were proven to be HIV infected at later time points. This was necessary as patients with such low positive results were previously shown to have a reduced probability of testing positive at a later time point [9, 10].
Pre-analytical DBS manipulation
Due to effective prevention of HIV mother to child transmission (PMTCT) programmes, the prevalence of HIV infection in young infants in South Africa is now low . With a very low prevalence the optimal theoretical pool size is large as most pools would remain negative, realising maximal cost savings for larger pools. Apart from the theoretical optimum we included other considerations: 1) the maximum number of DBS which could fit into a single reaction tube, 2) whether DBS could be added to a Roche S-tube directly or whether elution of DBS should be done as additional step without resulting in reaction inhibition (Roche specimen pre-extraction reagent [SPEX] was used as DBS eluant throughout), and 3) the sensitivity of pooled HIV PCR testing.
Pooled testing model
A model was developed to simulate yearly cost savings of a pooled testing approach compared to individual DBS testing. Daily sample and result data of the NHLS Medical Virology laboratory for the period 1 January 2009 to 31 July 2015 were used for the simulation (Additional file 1). The model assumed a minimum batch size of 10 samples for individual testing and 20 samples when pooling (i.e. between 4 and 10 pools at different pool sizes), to maintain a good turn-around time while maintaining batches of adequate size to justify the use of the instrument and laboratory personnel time. Remaining samples after minimum batch sizes were filled were added to the following day’s runs. For the observation period, the model simulated the number of positive pools when randomly allocating samples tested on a particular day to varying pool sizes from 2 to 5 with deconvolution of positive pools by individual testing, done the following day. This random allocation was repeated 1000 times, for each pool size, to estimate the mean values and variability in model output. Bootstrap confidence intervals around mean estimates are given by the 2.5th and 97.5th percentiles of the 1000 estimates.
Only the cost of reagents and consumables is considered. Labour costs and laboratory overheads, such as electricity and equipment costs, are not considered.
The model to estimate cost-efficiency of the pooling approach was implemented in the statistical software R  and a web-based tool was created using the Shiny package, accessible at: https://carivs.shinyapps.io/Calculator/. The user can enter average daily sample throughput, workflow management parameters, expected infant HIV positivity rate and reagent costs in US dollar. The model produces estimates of costs and batched runs saved at the optimal pool size, and a plot of cost savings as a function of pool size and positivity.