Identification of selected primary bloodstream infection pathogens in patients attending Kisii level five and Homa Bay county hospitals [version 2; peer review: 1 approved with reservations, 1 not approved]

Background: Bloodstream infection (BSI) contributes to a substantial proportion of mortality in sub-Saharan Africa and is marked by the presence of bacterial and/or fungal microorganisms in the blood. Because BSI can be life threatening, it requires a timely, reliable and accurate diagnosis. This study retrospectively analyzed data of identified BSI pathogens and compared the performance of the different diagnostic technologies used in terms of accuracy, sensitivity, turnaround time (TAT) and cost. Methods: Currently, culture followed by analytical profile index biochemical strips (API), (BioMerieux) are used as the conventional standard diagnostics in Kenyan public hospitals and labs. We compared the results of this standard to that of the BioFire FilmArray (FA) (BioFire Diagnostics) and MicroScan WalkAway-40 plus System (MS) (Beckman Coulter) used in diagnosis of BSI. The FA technology was able to identify 150/152 bacterial and yeast isolates with an overall accuracy of 99.04% (95% CI: 96.59-99.88%), (95% CI: 95.33-99.84%) and the mean TAT per sample was 53 and 103 hours for bacterial and yeast samples, respectively, with a running cost per sample of USD 28.05. Conclusions: The findings in this paper suggest that the FA and MS platforms should be able to perform adequately in Kenya referral hospitals and medical clinics as a rapid diagnostic tool. The study compare three methods of identification of pathogens growth in blood culture samples: API, Film Array and MS. All the 3 methods show excellent sensitivity and accuracy. What was the specificity?


Introduction
Bacteremia accounts for a large number of hospital admissions and results in high morbidity and mortality 1 . In sub-Saharan Africa, there is limited information on bloodstream infections (BSI) 2,3 , which can be partly attributed to the paucity of studies conducted in developing countries lacking high throughput BSI diagnostic technology. The availability of such equipment would facilitate widespread BSI detection in patients, help close this critical knowledge gap, and potentially save lives 4,5 .
Blood culture and analytical profile index (API, BioMerieux) strip analysis has been the conventional standard for bacteremia diagnosis in many hospitals throughout the world, including Kenya 6,7 . Using this technique has been a challenge because it is a labor-intensive process that requires experienced laboratory technologists and has a reportedly lower level of accuracy than other techniques such as conventional cultures [2][3][4][5][6] . To address this issue, diagnostic platforms such as the BioFire FilmArray (FA) and MicroScan WalkAway 40 plus (MS) are now widely used in Europe and the United States 2-8 .
The FA is a sophisticated closed-automated polymerase chain reaction (PCR) system that utilizes specific commercial pouches such as FA blood culture ID panel to identify BSI causative agents and antimicrobial resistance markers within 1 hour of a positive blood culture 2 . This platform can identify 24 causative agents of BSI (eight gram-positives, 11 gram-negatives, and give yeast species) and three antimicrobial resistance genes: mecA for methicillin, vanA/B for vancomycin and bla KPC for carbapenem by nested multiplex PCR 5-8 . The MS system can provide identification of bacteria within 16 to 20 hours, phenotypic antimicrobial profile (AST) and extendedspectrum beta lactamase (ESBL) using negative combo panel type 66 for gram negatives and positive combo panel type 39 for gram positives 9,10 and identification of yeast within 4 hours using a rapid yeast panel (Beckman Coulter, United States).
The use of conventional methods requires considerably long turnaround time (TAT) from 12-72 hours. In potentially life-threatening cases of BSI, the microbial cause of infection must be identified as quickly as possible to ensure proper treatment and management of the disease. The newly employed methods FA ® and MS ® have never been used before in Kisii and Homa Bay hospitals.
In this paper, we compared the accuracy, sensitivity, turnaround time (TAT) and cost of new technologies against the conventional API strip technique based on past data from blood culture isolates. Based on our findings, we will identify whether these automated platforms would be capable of providing a rapid diagnosis of BSI in Kenya hospitals.  The prevalence was taken from the study by Maze et al. (2018) 12 "The epidemiology of febrile illness in sub-Saharan Africa: implications for diagnosis and management". The prevalence rate is based on East Africa data which is part of Kenya, the ranges in prevalence is between 10-20% in Kenya.

Blood culture
Blood samples were collected into BACTEC Plus Aerobic/F, Peds Plus Aerobic/F, Anaerobic/F and Lytic/10 Anaerobic/F vials (BD, United States) and incubated using the BacTec 9050 instrument (BD, United States) for 5 days to account for slow-growing pathogens. A positive signal was indicated by an increased fluorescence caused by the carbon dioxide released by an organism reacting with the vial dye. Positive blood culture samples were removed and processed to identify the organism. Samples were processed directly using the FA without need for prior subculture. As for the MS and the API strip method, samples were first gram-stained and sub-cultured on MacConkey agar, Blood agar plate, Sabouraud Dextrose agar, Hektoen enteric agar and Tryptic soy agar (Becton Dickson).
Microbial identification of the positive blood culture samples For identification by FA, samples were tested per the manufacturer's instructions. First, the blood culture identification panel was inserted in the loading chamber, then hydration solution was added to the sample and then the panel was placed into the FA. The results were checked after 1 hour.
Pure colonies were used for the MS identification procedure. An inoculum of 0.5 McFarland standard equivalents was prepared by selecting 1 to 3 discrete colonies from pure culture on MacConkey agar (MAC) or blood agar plate (BAP) or Sabouraud dextrose agar (SDA) and suspended in 3 ml of the MS inoculum water (Beckman Coulter). From the solution, 100 μl was transferred and mixed with 25 ml of the MS inoculum water with pluronic and then poured into the sterile inoculator D set tray. The solution (140 μl) was transferred into a gram-negative or a gram-positive panel (Beckman Coulter), and then loaded into the MS and results checked after 18-24 hours and 4 hours for yeast organisms.
For manual biochemical analysis method, the API strips (BioMerieux, United States) and media (MacConkey agar, blood agar, Hektoen enteric agar, triple sugar irons and Sabouraud dextrose agar plates) were brought to room temperature. After this, 5 mL of deionized water was added to the tray along with the strips. The API ampules or equivalent suspension medium was inoculated with a single colony. The inoculation of wells for gram negative, gram positive and yeast organisms was done as per manufacturer's instruction. The incubation box was closed and incubated at 36°C for 18-24 hours for bacterial while for the yeast it was incubated at 29°C for 48 to 72 hours. A positive signal was indicated by a colorimetric change that was interpreted using the API guidelines.

Statistical analysis
Overall accuracy was calculated as the (number of individual isolate identified using evaluated technique) / (total number of same isolates identified) x 100%), sensitivity values and the 95% confidence intervals (CI) for both of these metrics were analyzed using the Graphpad Prism 8.2.1 software. The TAT was defined as the time taken from sample preparation to identification 13 . The average cost per samples included consumable reagents and disposable supplies and was defined as the total cost of each assay per sample run.
The indeterminate results were resolved by comparing the two techniques FA and MS, where the two techniques agreed was taken as a true positive, the discordant results were repeated using the API technique (gold standard).

Ethics approval and consent to participate
Ethical review for this work was obtained from the Kenya Medical Research Institute Scientific and Ethical review Unit-Scientific Steering Committee (KEMRI SERU-SSC) #3686 and Walter Reed Army Institute Research (WRAIR) Institutional Review Boards (IRBs) #2513. Consent was not sought for this study since it was determined that there was no interaction with human subjects.

Results
The sensitivity and accuracy of the FA, MS, and API The overall accuracy and sensitivity of each platform is shown in Table 1. For the FA, the calculated specificity was 99.04% (95% CI, 96.59-99.88%). Out of the total 152 isolates identified, FA technology was able to correctly identify 150 isolates resulting in a calculated sensitivity of 98.68% (95% CI: 95.33-99.84%). Similar to the specificity computation for the FA instrument, the specificity of the MS instrument was determined based on the number of isolates correctly identified by the MS out of the total number of isolates correctly identified. This yielded a specificity of 98.56% (95% CI: 95.86-99.70%). The MS technology identified 149 true isolates, which produced a calculated sensitivity of 98.68% (95% CI: 95.30-99.84%).  Table 2. Breakdown of pathogen identification by the FA, MS and API. The accuracy broken down by microorganism was calculated for each platform. The total number of isolates identified followed by accuracy percentage of FA, MS and API is listed above. For the FA, the "n/a" indicates that the isolates could not be identified beyond the family level (Enterobactericeae). Those rows with a dash mark (-) indicate those identified at specific genus level (Salmonella) and not group level (Enterobactericeae).  14 ).

The average cost of running one sample per technique
The total average cost of processing one sample using the FA technique was 140.11 USD ( Table 4). The total average cost of running one isolate per sample using the MS instrument was 38.75 USD while API technique was 29.17 USD. Extra equipment required but not included were biosafety cabinet, incubators, autoclaves, hot plates, conical flasks, stirrers and spatulas as well as the annual preventive maintenance for this equipment. Only the costs of the consumable items needed for each method were evaluated. As expected, the cost of the items used for the API technique was lower than for those used by the MS and FA.

Discussion
In resource-limited settings, the use of conventional methods in diagnosis of bacteremia has been a challenge to most public health facilities leading to misclassification of the diagnosis of BSI 8, 15 . The automated methods FA and MS proved to be more efficient, reliable and faster in the identification of a wide range of microorganisms than API. The technologies are reliable with a short turnaround time. These positive factors outweigh the use of API strips for microbial identification, which is considered the conventional standard in Kenya for diagnosis of BSI. In comparison to the FA and MS, the API method was more labor intensive. Furthermore, fastidious bacteria might not be identified if they fail to grow on culture media but can be identified directly from blood culture using FA.  species than automated platforms 20 . Interestingly, this did not occur with this study, and could partly be because the lab technician performing the assay had extensive clinical microbiology experience. Microbiology labs typically address this lower accuracy by adding biochemical tests such as oxidase and catalase to increase accuracy. We, however, did not incorporate these assays into the API strip analysis.
The mean TAT difference per run of eight samples among the technologies was significant at p<0.0001. The FA technology required 8 hours 48 minutes per eight samples compared to the MS, which required 42 hours and the API method, which required 53 hours for bacterial species and 103 hours for yeast. The major factor contributing this difference was time needed to prepare the isolates, which require gram-staining then culturing for 24/48 hours prior to identification by MS or API 20 . It should be noted that the MS has higher throughput and can process 40 or more panels in one run. In addition, the API method can test more samples and is only dependent on availability of incubators, reagents and the experience of the technician. The shorter TAT for FA is a very attractive feature for under-developed areas with poor infrastructure and inaccessible areas where field clinical/research activities are undertaken and do not necessarily require a high-throughput machine. Though not a metric evaluated in this study, the FA requires considerably less training and skill compared to the other methods, which help to balance its throughput limitations.
The average cost of testing one sample using FA was noticeably higher than the cost of the MS and API methods. This was expected as the FA test kits cost more than the MS panels and the API reagents. While the FA is more expensive, it is able to identify co-infection in one sample, which would require separate runs for the MS and API 5 .
Of note, the FA is able to identify resistant genes such as methicillin resistance common with Staphylococcus aureus, vancomycin resistance common with Enterococcus spp. and carbapenem resistance common with Klebsiella pneumoniae and other Enterobactericiae 8 . While the MS has no capability to identify antimicrobial resistant genes commonly associated with BSI, it is able to perform phenotypic drug sensitivity. In fact, the MS has a wider range of antimicrobial testing capabilities with regularly updated software database in line with CLSI guidelines.

Conclusion
While the evaluated methods were similar in accuracy and sensitivity, there were appreciable differences in TAT and cost. The FA cost more, but had a quicker TAT compared to the MS and API methods. This is a significant concern when using the machine in areas with limited financial resources. However, the FA requires minimal training prior to use and is able to identify co-infections. Furthermore, the FA requires a small space, and therefore, the cost of the FA panels should not be considered a major drawback since early detection of BSI has shown to reduce medical costs, hospital stays, and help guide In past studies, the API strip analysis had a lower accuracy identifying microorganisms such as Citrobacter species, Escherichia coli, Pseudomonas aeruginosa and Enterobacter the clinicians on the best treatment approach 5 , which lower overall economic costs.

Recommendation
The FA and MS have not been evaluated at Kenya hospitals and further evaluation using a larger sample size is recommended in order to have more data on BSI pathogens and their antimicrobial susceptibilities in different localities. However, these preliminary results clearly suggest that both the FA and MS platforms are valuable tools in rapid identification of BSI. Each technology has its advantages and disadvantages, which must be considered. Still, implementation of either platform could result in reduction of hospital stays, lower cost, better patient management and more appropriate use of antibiotics by clinicians. to the family level, following manufacturers' recommendations and inhouse standard procedures this can be identified further to genus level using salmonella antisera. The FA identification panel keeps on improving and the current panel (BCID2) identifies Salmonella at genus level using the same machine/platform. The turnaround time for FA is really helpful in early diagnosis of Salmonella.

Data availability
In the discussion, the authors state that S. anginosus and S. bovis are contaminant. I do not agree since these Streptococcus may be responsible of endocarditis or other (abdominal) infections.
Response: Thank you for pointing this out, These microorganisms cause bacteremia taking advantage of immunocompromised patients though not common in immunocompetent individuals.
The identification of the bacteria is not enough. More data on antimicrobial susceptibilities are required since antimicrobial resistance is a major problem in Africa. So, more data are needed.