Aetiology and severity of childhood pneumonia in primary care in Malawi: a cohort study

Objective To determine the aetiology of community acquired pneumonia in children presenting to primary care in Northern Malawi, and to ascertain predictors for identification of children requiring hospitalisation. Design The BIOmarkers TO diagnose PnEumonia study was a prospective cohort study conducted from March to June 2016. Setting Primary care in Northern Malawi. Patients 494 children aged 2 –59 months with WHO defined pneumonia. Main outcome(s) and measure(s) Number of children with bacterial infection identified and the sensitivity/specificity of WHO markers of severity for need for hospitalisation. Results 13 (2.6%) children had a bacterium consistent with pneumonia identified. A virus consistent with pneumonia was identified in in 448 (90.7%) of children. 56 children were admitted to hospital and two children died within 30 days. 442 (89.5%) received antibiotic therapy. Eleven children (2.6%) had HIV. WHO severity markers at baseline demonstrated poor sensitivity for the need for hospitalisation with a sensitivity of 0.303 (95% CI 0.188 to 0.441) and a specificity 0.9 (95% CI 0.868 to 0.926). A prediction rule to indicate the need for hospitalisation was developed. Conclusions and relevance The low rate of bacterial infection and high use of antibiotics in the setting of high immunisation rates highlights the changing profile of childhood pneumonia. Similarly, the markers of need for hospitalisation may have changed in the setting of extended immunisation. Further studies are required to examine this.

the current WHO markers of severity combined clinical symptoms, signs and oxygen saturation measured by pulse oxymeter 5. Page 6, line 50 -57: I would recommend more information on the study site and population. For example, more of the health facilities and the services available. Also, a little more on the population these facilities serves (hence the population from which the study group is drawn) 6. page 8 line 42 -50: I recommend more information on how samples were processed. For instance for blood cultures what procedure was used? for the PCR, which PCR method was used? 7. Page 10, line 21 -23: I need to understand why there was a follow-up period in this study? 8. Page 10, sub-heading (immunisations): I do not see the importance of this result presented. I recommend comparing immunization rates between hospitalized and nonhospitalized cases (as bacterial pneumonia are more likely to be severe and hence patients hospitalized). 9. Page 10, line 47 -59: it is not clear the source (i.e sample source) from which these bacteria/viruses were isolated? (Blood?NPS or both?) also the method of identification should be clarified. (PCR, culture, both) 10. Page 10, line 54: it will be difficult to talk of causality based on NP swabs alone as this cannot discriminate carriage (therefore is not specific for disease). I see this is touched on in the discussion section. Consider changing the wording of this sentence. 11. Page 12, line 18: This statement is likely linked to reference No. 10 (Hammit et al) and not No. 6. 12. Page 12, line 45 -49: The markers of severity you proposed include symptoms/features that do not discriminate severe pneumonia from other illnesses (eg stridor, head nodding, convulsion etc).

REVIEWER
Edem, Bassey MRC Unit The Gambia at LSHTM REVIEW RETURNED 10-Jan-2021

GENERAL COMMENTS
The Authors have demonstrated evidence of the changing aetiology of childhood pneumonia in Malawi post introduction of conjugate vaccination and also developed a tool for predicting hospitalization that can be validated in larger studies. However, it is not mentioned at any point how the sample size for this study was estimated. This limits the external validity of the study findings as a whole and speaks to the generalizability of the study findings (lines 47-50).
As a background to the study, it would be useful to provide coverage estimates for Malawi for PCV and HiB routine immunization. 5. Page 6, line 50 -57: I would recommend more information on the study site and population. For example, more of the health facilities and the services available. Also, a little more on the population these facilities serves (hence the population from which the study group is drawn) We have added further details to the main manuscript and a supplemental file due to space considerations. The detail in the main manuscript is "Mapale Health Centre currently serves the whole city as a primary health care facility (population of approximately 220,000 ) The under five population in this setting is characterised by a high burden of infectious disease including acute respiratory infections (ARI), diarhorrea, and malaria. In 2016 the estimated coverage was 83% for pneumococcal conjugate vaccine and 84% for Haemophilus influenza type B vaccine in Malawi, 12 (Further details on site are available in the supplemental file)" The supplemental file contains Mzuzu is one of the four cities in Malawi, with a population of 221,272 (National Statistics Office, 2018). Mzuzu health centre also known as Mapale or Mzimba North Health centre is stituated at the City centre of Mzuzu city while Mzuzu Central Hospital is situated at around 4 km Northwards of the city. The health centre currently serves almost the whole city as a primary health care point while the central hospital is regional referral tertiary hospital serving six districts in the northern region, over hundred health centres and serving a population of over 1.9 million people. It also serves as a secondary healthcare facility for the local (Mzuzu city) population. The majority (98.0%) of our participants were from Mapale health centre. The Central hospital initially served as a primary healthcare facility also but prior to commencement of the study the local health authority decided to maintain it as a secondary care facility only. Therefore it introduced a "bypass fee" where patients who attended the hospital clinics without being referred from a health centre/district hospital were charged MK 1,500 (appromately 2 Euros). This led to Mapale becoming the main site of recruitment due to the focus on recruiting patients from primary care and before they had received any treatment. This study was conducted in primary care clinics. Patients who required admission were referred to hospital in a different location and were not admitted to the clinic where the study was undertaken. In order to ensure that any patients who may have went home and were subsequently presented to another facility or went directly to hospital were identified we subsequently tried to contact caregivers to identify those patients who were admitted to hospital following their first presentation by mobile phone follow up.
8. Page 10, sub-heading (immunisations): I do not see the importance of this result presented. I recommend comparing immunization rates between hospitalized and non-hospitalized cases (as bacterial pneumonia are more likely to be severe and hence patients hospitalized). We agree that this is a useful analysis and we have included this now in the revised draft. In order to take account of age differences we also undertook further analysis also (text in manuscript below) Furthermore, of the 56 patients who were hospitalised, 45 had immunization records available. We age and sex matched this group in a 1:2 ratio with 90 patients who were not-hospitalised and had immunization records available (Supplemental file). The median number of immunisations amongst the hospitalized cohort was significantly lower than the non-hospitalised cohort ( We have now included the details below in the statistical analysis section: The overall aim of this project was to determine blood and urine biomarkers of host response in bacterial pneumonia in a cohort of children age 2-59 months in a single centre in Malawi. Based on the study by D'Acrement 6 et al it was estimated that the prevalence of bacterial infecion among febrile children with pneumonia would be 20%. The required sample size for our study based on this prevalence was given by the formula n= 4pq/d 2 where p = prevalence q = 100-p and d= the precision of the estimate. Assuming a relative precision of 20% and based on previous pilot work on follow up and sample analysis this resulted in a required sample size of n= 490.
As a background to the study, it would be useful to provide coverage estimates for Malawi for PCV and HiB routine immunization.
We have added the following text: In 2016

GENERAL COMMENTS
Thank you for the revised manuscript, and for addressing all my comments. I do not have any specific comment, rather I have a more general comment regarding your proposed severity score. I think your severity score in itself is interesting, though it does not appear to add much to the current WHO integrated management of childhood illness (IMCI) criteria for pneumonia severity. Importantly, by including a severity grade, your proposed severity score now sets a higher threshold for definition of severity than the IMCI guide (which defines a child as having severe pneumonia if they have at least one sign or symptom of severity). I am also concerned about how you attempted to validate your scoring system, using "hospitalisation" as a sort of reference for severity. This makes it a bit problematic because in practice, the decision to admit a child with pneumonia is not based entirely on the severity of their symptoms. Specifically, in line 15 -21 of page 13, you report the sensitivity and specificity of the scoring system against hospitalizations. I would suggest reporting on associations between scores and hospitalization probability, and report sensitivity and specificity against a more widely accepted measure of severity like the IMCI criteria.

VERSION 2 -AUTHOR RESPONSE
I do not have any specific comment, rather I have a more general comment regarding your proposed severity score. I think your severity score in itself is interesting, though it does not appear to add much to the current WHO integrated management of childhood illness (IMCI) criteria for pneumonia severity. Importantly, by including a severity grade, your proposed severity score now sets a higher threshold for definition of severity than the IMCI guide (which defines a child as having severe pneumonia if they have at least one sign or symptom of severity).
These are useful comments to refine the discussion. We accept that the use of a severity score rather than a single item score such as WHO may make calculation more difficult and requires development of a threshold at which referral is indicated. However there are a limited number of items and such scores are widely used in clinical practice already [1] . We also believe that the score does add value compared to the WHO score. In our study the WHO signs had high specificity but poor sensitivity. In primary care we believe that a high sensitivity score is required to ensure that false negatives are limited and that children are referred to hospital and that a high sensitivity score is more appropriate in this setting [2] It also concords with the concerns raised in the study by Agweyu et al. [3] showing that 39% of fatal cases of pneumonia were defined as having non-severe pneumonia requiring only home treatment by the 2013 WHO revision We have added the following to the text to This score is predictive of hospitalisation, independent of age, malaria status, HIV status, number of vaccinations, number of people usually sleeping in the same room as the child, presence of chimney for indoor fire, availability of electricity in the home, distance from the nearest health clinic and staff designation of severe pneumonia using WHO IMCI criteria. The DIFFICULTY DRAWING breath score was also independently associated with staff designation of severe pneumonia. However, we observed that there was no relationship between staff designation of severe pneumonia using WHO criteria and the decision to admit. This score does require added calculation compared to the WHO severity markers which only require any one of the items and also requires a decision threshold as to the level that requires referral to hospital. However multicomponent scores are common in clinical practice and we believe that evaluation of such scores and consideration of trials integrated them into electronic decision supports is warranted. The score here also has a high sensitivity compared to the WHO score which has a high specificity. We believe that in primary care a score with a high sensitivity is more appropriate to ensure that false negatives are limited. These aspects will need to be evaluated and validated in future studies.
I am also concerned about how you attempted to validate your scoring system, using "hospitalisation" as a sort of reference for severity. This makes it a bit problematic because in practice, the decision to admit a child with pneumonia is not based entirely on the severity of their symptoms. Specifically, in line 15 -21 of page 13, you report the sensitivity and specificity of the scoring system against hospitalizations. I would suggest reporting on associations between scores and hospitalization probability, and report sensitivity and specificity against a more widely accepted measure of severity like the IMCI criteria.
We accept that hospitalisation is a subjective standard and may vary based on hospital practice and health seeking behaviour. However since this study is based in primary care we believe that hospitalisation is a useful measure of severity. WHO markers of severity are used to guide who should be referred to hospital from primary care also We have added the following piece to our limitations section "The use of hospitalisation as a marker of severity may be seen as a limitation as hospitalisation may vary based on hospital practice and health seeking behaviour. However this study was based in primary care and hospitalisation is a useful standard to guide those who should be referred from primary care. Larger studies to evaluate and validate scores for markers such as mortality are required" We have also undertaken further analysis to control for co-variates which is presented in the supplemental file for both hospitalisation and WHO defined severe pneumonia S3: Markers of severity Figure S1 Difficulty DRAWING Breath score and risk of hospitalisation DIFFICULTY DRAWING breath score = 1 for each of: Missing Patients0 [1] The DD Score (strongly, positively associated) and previous number of vaccines (modestly, negatively associated) are the only independent predictors of hospitalisation in the model with the following covariates included: • For each unit increase in the DD Score there is a 10.6% (95% CI 7.9-13.4) increase in the likelihood of severe disease as defined by hospitalisation (p<0.0001)