Admissions to a low resource neonatal unit in Malawi using The NeoTree application: A digital perinatal outcome audit

Background: Mobile-health has increasing potential to address health outcomes in under-resourced settings as smart-phone coverage increases. The NeoTree is a mobile-health application co-developed in Malawi to improve the quality of newborn care at the point of admission to neonatal units. While collecting vital demographic and clinical data this interactive platform provides clinical decision-support, and training for the end-users (health care workers (HCW)), according to evidence based national and international guidelines. Objective: Our aims were to examine one month of data collected using the NeoTree in an outcome audit of babies admitted to a district-level neonatal nursery in Malawi and to demonstrate proof of concept of digital audit data in this setting. Methods: Using a phased approach over one month (21 Nov – 19 Dec, 2016), frontline HCWs were trained and supported to use the NeoTree to admit newborns. Discharge data were collected by the research team using a discharge form within the NeoTree ‘NeoDischarge’. Descriptive analysis was conducted on the exported pseudonomysed data and presented to the newborn care department as a digital audit. Results: Of 191 total admissions, 134 (70%) admissions were completed using the NeoTree and 129 (67%) were exported and analysed. Of these 129, 102 (79%) were discharged alive. Overall case fatality rate was 93 per 1000 admitted babies. Prematurity with respiratory distress syndrome, Birth Asphyxia, and Neonatal sepsis contributed to 41.6%, 58.3% and 16.6% of deaths respectively. Deaths may have been under-reported due to phased implementation and some families of babies with imminent deaths self-discharging home. Detailed characterisation of the data enabled departmental discussion of modifiable factors for quality improvement, for example improved thermoregulation of infants. Conclusions: This digital outcome audit demonstrates that data can be captured digitally at the bedside by HCWs in underresourced newborn facilities and these data can contribute to meaningful review of quality of care/outcomes and potential modifiable factors. Coverage may be improved during future implementation by streamlining the admission process to be solely via digital format. Our results present a new methodology for newborn audit in low-resource settings and are a proof of concept for a novel newborn data system in these settings. (JMIR Preprints 03/10/2019:16485) DOI: https://doi.org/10.2196/preprints.16485


Table of Contents
developed in Malawi to improve the quality of newborn care at the point of admission to neonatal units. While collecting vital demographic and clinical data this interactive platform provides clinical decision-support, and training for the end-users (health care professionals (HCP)), according to evidence based national and international guidelines. Objective: Our aims were to examine one month of data collected using the NeoTree in an outcome audit of babies admitted to a district-level neonatal nursery in Malawi and to demonstrate proof of concept of digital outcome audit data in this setting. Methods: Using a phased approach over one month (21 Nov -19 Dec, 2016), frontline HCPs were trained and supported to use the NeoTree to admit newborns. Discharge data were collected by the research team using a discharge form within the NeoTree 'NeoDischarge'. Descriptive analysis was conducted on the exported pseudonomysed data and presented to the newborn care department as a digital outcome audit.

Introduction
2.5 million newborns die each year with no registration of their death or any documentation of how or why they died [1]. Half of the world's newborn babies do not receive a birth certificate [2,3] and stillbirths are statistically invisible [3]. Despite this, it is widely acknowledged that in order to prevent newborn deaths, more information on numbers of births and deaths, causes of deaths and avoidable factors linked to deaths is needed [4]. The process of perinatal death audit aims to establish the profile of facility-based causes of death and has been shown to reduce perinatal mortality by thirty percent in low resource countries [5] . Improving quality of care for newborns is a key priority in tackling persistently high neonatal mortality rates (NMR) particularly for hospitalised sick newborns in lowand middle-income countries [6]. At the facility level the process of audit and feedback is considered the cornerstone of quality improvement, particularly when the process includes a clear action plan and targets [4]. Meanwhile, smart phone technology is becoming increasingly common place in lowresource settings. M-health ("the use of mobile and wireless technologies to support the achievement of health objectives" [7]) has been harnessed for accurate and efficient data collection particularly in community settings and in the context of clinical trials [8,9]. Thus far, to our knowledge, m-health has not been utilised in the context of hospital perinatal death audit. In response to recognised challenges in the scale up of audit in these settings and a call for electronic health systems [ 5] we present the results of a novel digital outcome audit collected by health care professionals (HCPs) on a m-health application: the NeoTree [10].
Types of clinical audit exist for different purposes, for example a structural audit examines the availability of resources in a system and a process audit assesses the process of case management [5]. An outcome audit assesses the end results of care; either deaths or near misses depending on the NMR. In high income countries where NMR is low, the focus is more on near-misses and the management of specific cases are typically discussed in a monthly multidisciplinary 'morbidity and mortality meeting'. In low resource settings where NMR is high the emphasis is on deaths and these are often discussed in a 'death-audit' meeting. Perinatal death audit has been defined as "the process of capturing information on the number and causes of all still births and neonatal deaths, or near misses where applicable, with an aim toward identifying specific cases for systematic, critical analysis of the quality of perinatal care received in order to improve the care provided to all mothers and babies" [4]. It aims to follow a six-step cycle summarised in Figure 1  Despite being one of the least developed countries in the world Malawi has seen great success in achieving Millennium Development Goal four [11] particularly for under five mortality, however for newborns, mortality is persistently high. A previous paediatric death audit in Kamuzu Central Hospital (KCH), a large referral centre in Malawi, was recently published and cited reliable record keeping as a significant limitation with six percent of charts missing and documentation deficiencies in 58 percent of the charts reviewed [12]. To our knowledge a perinatal death audit has not yet been published from Malawi or other sub-Saharan countries.
Our aim was to conduct and report a digital outcome audit of admission and discharge information collected by HCPs in a district level facility using a novel m-health application the NeoTree on electronic mobile devices.

Setting
Zomba Central Hospital (ZCH) is a district level hospital in Malawi with a large neonatal unit. During our study period and in discussion with the local clinical management team it was noted by the researcher-in-residence (CC) that; the 40-bed nursery takes around seven admissions per day with nine staff covering day and night shifts on three neonatal wards (High dependency, transit and kangaroo wards). At maximum capacity, this equates to up to three nurses looking after approximately 50 babies per shift. Admissions are referred from a range of areas, defined as their "place of origin", from within ZCH (e.g. Theatre), or from outside ZCH (e.g. home, health centre, or other hospital). A health centre in this context is a facility that provides outpatient care services for common diseases in the local population, while a hospital is a larger facility providing more specialised care to a district population. Admissions are usually clerked on a Malawian Ministry of Health (MOH) paper proforma. Each patient's paper medical record consists of this admission form and any other loose-leaf sheets held together with string. The records of Neonatal Deaths (NNDs) are examined by paediatric/neonatal HCPs at monthly death audit meetings, often without obstetric input. A rudimentary root cause analysis for each death is postulated and subsequent recommendations made. It was observed informally by the researcher-in-residence [CC] that the quality of the paper records is very poor [12], with faded, illegible writing, and the death audit process is time-consuming, often taking a whole day. She also observed that clinicians find it particularly difficult to attend death audits as their clinical duties continue. Oxygen concentrators and heaters in the unit rely on an electricity supply which is affected by power outages on a daily basis. When these occur a backup supply of electricity is provided by a generator within a few minutes.

Identifying deaths and collecting information using a digital app (Stages 1 & 2 of audit cycle) The NeoTree App
A bedside application; the NeoTree, was used (in addition to MOH paper admission form) by frontline neonatal staff (mainly nurses) to record one month of admissions (21 Nov-19 Dec, 2016) to the neonatal unit on three low-cost android electronic tablet devices, which were provided and installed at the nursing station. This audit took place in the context of an intervention development study [10], which presents the co-development process and acceptability, feasibility and usability data.
The NeoTree app was structured as an electronic admission form ( Figure 2) which included all the fields of the standardized MOH paper neonatal admission form. Due to HCP feedback some fields were removed (e.g. physical gestational score) but no new fields were added. By the end of the study period the digital form followed the structure as below.
Place of origin 6.

Digital admission form
Most fields on the electronic admission form were compulsory in that the HCP-users could not continue until a plausible value had been recorded. However, in attempts to make the form more user friendly and practical, some fields that relied on availability of specific equipment (such as blood sugars and oxygen saturations) were made optional. Data collected at admission and discharge were stored locally on the tablets and a printed copy was kept in each patient's paper medical record which included patient identifiers. When a network connection became available data were exported pseudoanonymised with a unique ID to the researcher-in-residence's (CC) laptop as an excel file.
Fields containing identifiable information were configured as confidential by the researcher-inresidence so that they were not exported.
During each digital admission, following emergency triage and stabilisation of the baby, a 'reason for admission' was recorded. This was the presenting complaint entered by the nurse according to what the baby was referred with from labour ward or other facility. If the referred patient arrived without accompanying information, HCPs had to use their own clinical judgement. This may have differed considerably from the actual diagnosis, where for example; labour ward referrals were labelled 'Meconium aspiration' simply because there was meconium at delivery. Reason for admission is mutually exclusive, as there was only one presenting complaint recorded for each baby. Later at the end of the digital admission form, 'provisional HCP admission diagnoses' were decided by the HCPs based on their initial assessment of the baby. They could choose more than one diagnosis when necessary, hence these were not mutually exclusive.
During admission, gestation was estimated from fundal height (recorded antenatally) and length of pregnancy (reported by the mother), both of which are unreliable methods [13,14], hence a maturity score was included in early iterations of the NeoTree (Appendix 1). Occasionally wireless printing of forms was momentarily delayed by power outages but these did not affect completion of forms on the tablets since they had a 6-7 hour battery life.

Digital discharge form
The discharge form, the 'NeoDischarge' was completed by a second researcher (EK) at the point of discharge, when the baby died or was taken home by their mother against medical advice (i.e. absconded). Discharge information included identifiable information and outcome (discharged alive, absconded or died), HCP discharge diagnoses 1, 2 and 3 for discharges, and cause of death for NNDs. Often there was more than one HCP discharge diagnoses documented hence the need for three fields and these data were not mutually exclusive. It was noted that some patients in the sample had not been reviewed by a clinician during admission hence the researcher-in-residence (CC, a UK paediatrician in training with 1.5 years experience in LIC newborn care), decided the most salient singular discharge diagnosis for all babies. This was labelled 'researcher in residence discharge diagnosis' as shown in Table 2.

Recruitment & initial training
All nine-permanent staff on the neonatal rota were invited to attend a scenario based one-on-one session in which they were trained to record new admissions using the app. Their written consent to take part in the study was obtained according to ethical approval from The Malawi College of Medicine Research and Ethics Committee (COMREC; Ref:17PP12). Each HCP was then supervised clerking (recording) one new admission on the ward, completing both MOH paper form and the NeoTree app. When patients arrived with partially completed paper forms HCPs were encouraged to cross-check information already documented, with mothers and the mother's medical record-book, before entering it to the app. Permanent staff who could not attend the initial training session gave written consent when they entered the study (n=1). The audit and implementation was explained to other ad-hoc locum staff or nursing students by the head nurse of each shift, who obtained their verbal consent.

Phased implementation and iterative changes
The NeoTree was then implemented on the ward in a phased approach over one month, with increasing coverage and decreasing levels of supervision. In the first week of the study, day-time admissions were supervised by the researcher-in-residence (CC) who was present on the ward. She was also available throughout the one month study period to support any technical issues that occurred. Electronically clerked admissions were cross-checked by the researcher-in-residence with the admissions book and the ward manager at the end of each day to see if any patients had not been captured on the NeoTree. Incomplete sessions were deleted. During the study, via an on-line editor platform the researcher-in-residence was able to act on verbal feedback from the nurses as they used the app, and re-configure the NeoTree, to best fit an efficient admission process, in consultation with the wider NeoTree team. For example, the order of sections was adjusted to start with triage and examination and end with mothers' history (the opposite of the order of the MOH paper form), and pictures were added to explain exactly how to check danger signs.

Statistical analysis (Stage 3 of audit cycle)
The primary outcome of this study was to report and describe deaths over a one-month period using an electronic app NeoTree. To this end the admission and outcome data sets were exported from the app and merged by matching unique identification numbers. The SPSS programme was then used to analyse the data and produce simple graphs [15]. Descriptive statistics included totals (n), percentages (%), mean and standard deviation (SD) for normally distributed data as well as median and interquartile range (IQR) for skewed data and ranges. Charts included simple bar and pie charts for categorical data and histograms for continuous data (e.g. a histogram for temperature). To measure the digital process, the number of missing pieces of data for each data point were reported as a total (n) and as a percentage(%). The average time taken to complete the app was also calculated. Case Fatality Rates (CFRs) for each diagnosis were calculated as the number of deaths per 1000 babies admitted with that diagnosis e.g. CFR for Respiratory Distress syndrome (RDS) was 148 deaths per 1000 cases of RDS. The overall CFR was calculated as the number of deaths per 1000 babies admitted.

Presentation of results at an audit meeting (Stage 4 of audit cycle)
Simple graphs and statistics produced were presented by the researcher-in-residence (CC) to the department two weeks later in conjunction with that month's death audit meeting. The researcher facilitated the discussion regarding each graph sequentially, however the discussion was led by the participants. Emerging patterns in admission and outcome data, case fatality rates, and modifiable factors contributing to morbidity and mortality were identified and possible solutions were discussed. Individual cases were not discussed.

Participants
A total of 31 HCPs of four different cadres clerked newborns using the NeoTree app, including all nine-permanent staff on the rota, eight of whom attended the one-to-one training workshops at the beginning of the study. All other 22 participants were non-permanent locum staff or nursing students temporarily working on the unit (see Table 1).  Participants attending the audit meeting included three female nurse midwife technicians, one male clinical officer (who was also the head of department) and one male nursing officer.

Digital outcome audit findings (Stage 3 audit cycle)
During the study period there were 191 admissions to the neonatal ward. A total of 134 admissions were completed using the NeoTree (70%) and data from 130 babies were exported for analysis. Of these, 129 were analysed due to one repeat entry. Hence 129/191(67%) of all admissions were analysed. Table 2 describes patient demographics and clinical data. Key highlights are described here. Mean birth weight was 2616g with just over one third (34%) of babies born low birth weight and 45% born prematurely at <37 weeks' gestation. In total 25 maturity scores were carried out by HCPs most of which estimated higher gestation than the fundal height or length of pregnancy methods (data not shown).  Mutually exclusive = Assumes one single diagnosis only b Not mutually exclusive = Assumes on discharge some babies had more than one diagnosis which may have contributed to their presentation Figure 3 depicts the clinical reason for admission, recorded at the beginning of the app before clinical assessment, the most common of which was fever (30%) followed by birth asphyxia (17%) and prematurity (14%). The provisional admission diagnoses, made by HCPs (at the end of the app, therefore after clinical assessment) are reported in Figure 4. The most common discharge diagnoses recorded by the HCP and the researcher-in-residence were very similar ( Table 2): sepsis followed by birth asphyxia and low birth weight. In total, 102 babies (79%) were discharged alive, 7 (5.4%) left the hospital against medical advice and 12 (9.2%) died. Table 3 describes clinical findings at emergency triage and HIV status of the babies. Two thirds of babies were not crying at triage and hence underwent an ABCD (Airway, Breathing, Circulation, Disability) examination with only a minority considered 'unstable' following this assessment. Grunting was the most common danger sign in the unstable babies. Over a third (38%) of neonates admitted were hypothermic. A fifth of babies were known to be exposed to HIV before or during birth. Cause of death for the 12 newborn deaths, as recorded by the second researcher (EK) on the discharge form, are shown in Table 4. In all 12 patients cause of death was the same as the researcher in residence discharge diagnosis. Prematurity with RDS and Birth asphyxia were the leading causes of death.

Examination of the newborn:
A summary of findings on examination of the newborn are detailed in Appendix 2. 41 admissions (32%) had signs of difficulty breathing or respiratory distress where chest in-drawings was the most commonly reported sign (28%). 12(9%) were deemed to have severe work of breathing by clinical judgement. 56(43%) of HCPs reported confidence in using a stethoscope, however there was a paucity of chest findings reported and no heart murmurs were reported at all. Examination of fontanelle was unavailable in the data because the confidential button had been pressed in error, by the researcher-in-residence during the configuration of this field.

Place of origin:
In terms of place of birth; 93(72%) of the admissions were born in hospital, 3(2.3%) were born at home, 27(21%) in a health centre (HC), 4(3%) were born before arrival(BBA). 2(1.6%) were born with a traditional birth attendant despite a 'ban' of traditional birth attendants by the government. 40 patients were referred from other facilities (Appendices 3 and 4).

Maternal and Antenatal History:
Age of mother was poorly recorded as 62 out of 129 mothers (48%) did not know their exact date of birth (DOB) and was not exported for analysis since it is potentially identifiable information. A summary of the maternal history captured by the app is shown in Appendix 5. Attendance at antenatal care was generally poor (Appendix 6) with 31% attending 2 or less antenatal appointments. A minimum of 3 antenatal appointments are needed to receive all doses of tetanus vaccine. Most mothers (98%) had been tested for HIV and 25(19%) of those tested had a positive result ( Figure 5). Two of the three HIV exposed babies that did not receive nevirapine prophylaxis, had mothers who delayed Highly Active Anti-retroviral Therapy (HAART) until the second trimester, hence were the most vulnerable to vertical transmission of HIV. Syphilis status was much more poorly recorded in comparison to HIV status, with 'unknown' in 21 mothers (16%). 35 mothers (27%) had definitely not had a syphilis test. Of 73(57%) mothers who had been tested, three were positive and all of their babies were treated with penicillin. Medical conditions in pregnancy included malaria (16 mothers (12%)) hypertension(2), other sexually transmitted diseases(2) and anaemia (2). No maternal heart disease, diabetes, or thyroid disease was reported in the NeoTree. Eight mothers were administered antenatal steroids (6.2%).

Labour History:
A summary of the data captured for labour history is shown in Appendices 7 and 8. Maternal conditions reported in the labour field of the NeoTree app included significant vaginal bleeding seen in four mothers but no other problems were reported.

Completeness of data
Most fields, in the app were compulsory hence data for these fields were 100% complete with no missing data. For example, respiratory rate (RR) and temperature (T) were recorded for all 129 babies in the sample (these were compulsory fields: Table 2). Six fields were not compulsory and therefore had less than 100% data completion rates; admission HR (94%) and admission oxygen saturations (94%), birth weight (91%), blood sugar (96%), saturations in oxygen (19%) and outcome (94%), The seven missing admission HR and saturations readings occurred when a pulse oximeter had not been available. In these cases, the app directed the HCP to take and record a manual HR, hence some form of HR (electronic or manual) was recorded for every baby. HCPs reported that birth weight was not always available for older infants born in other facilities and that missing blood sugar readings occurred when strips for the blood sugar monitor were not available. Saturations in oxygen was only supposed to be recorded when babies required oxygen. The eight missing outcomes were due to not outcome being recorded or illegible handwriting in the patient notes.

Time taken
The mean time taken to complete an admission using the NeoTree app was 37 minutes (range 18-59 minutes), excluding one outlier (n=1). Anecdotal reports form HCPs suggests longer sessions may have been interrupted by urgent tasks. Approximately one fifth (22%) of NeoTree admissions were supervised.

Emerging patterns and corresponding modifiable factors discussed (Stage 4 of the audit cycle):
The audit meeting took just over one hour, after which HCPs could return to their daily duties. Four factors were discussed; first, the high rate of hypothermia on admission ( Table 3) and thermoregulation of babies by drying and wrapping was identified as a modifiable factor for improvement and future re-auditing. Hypothermia is also an example of a factor that could be highlighted in the anticipated next phase of NeoTree development, i.e. feedback data-dash boards linked to the NeoTree data. Second, inadequate reporting of antenatal syphilis testing was discussed, and it was suggested this could be fed back to health centres via the District Health Officer. Furthermore, the timing and completeness of penicillin treatment was requested to be added to the NeoTree form. Third, difficulties in knowing mothers' age was also highlighted as important due to younger mothers typically experiencing more premature deliveries and more complications of pregnancy. A request was made for the app to calculate this in the future. Finally, due to the lack of reported problems in labour other than bleeding, it was deliberated that this question may have been answered poorly and that midwives and newborn HCPs should be shown where to find this information.

Discussion
This paper presents a novel approach to capturing documentation in the inpatient setting which could signal a start to inpatient computerisation. To our knowledge this is the first digital outcome audit of neonatal admissions to a low resource newborn facility. Our digital outcome audit achieved 70% coverage of admissions during a phased implementation approach where HCPs collected the data themselves on a novel app, the NeoTree [10]. The overall case fatality rate for newborns admitted on the NeoTree was 92 per 1,000. The commonest diagnoses were sepsis, prematurity and birth asphyxia for which the case fatality rates were 34, 250 and 250 per 1,000 respectively. Completeness of data was high or 100% for much of the data set, exemplifying how the digital method significantly improves quality of data in terms of completeness. In comparison, other studies have commented how >50% charts had missing documentation [12]. Our one-month audit has completed steps one to four of the audit cycle ( Figure 1 [5]) and has the potential for re-audit, evaluation and refinement of recommendations and hence completion of the whole audit cycle.
In a previous study at ZCH nursery, demographic data were collected over a 2-month period using the MOH paper admission form. The mortality rate was 160 per 1,000, significantly higher than our digital outcome audit [17]. Therefore our digital outcome audit could have underestimated case fatality rates due to systematically missing data on babies who died. As HCPs were required to complete a paper form in addition to the NeoTree, they may not have filled out a NeoTree admission for babies that died soon after birth. If the NeoTree completely replaced the paper option and HCPs were trained to clerk all babies, including those arriving moribund, this could cease to be a problem.
Another previous paper-based death audit in KCH hospital Malawi audited paediatric patients, with ages ranging from 1 day to 16.5 years, rather than newborns and showed mortality rates ranging from 22 to 44 per 1,000 [12]. The lower mortality rates most likely reflect the older age range but may also be due to the retrospective nature of their study and missing data. The authors reported that >50% charts had missing documentation [12]. The prospective nature of our study, the presence of a researcher onsite overseeing data collection and the use of a digital method may have aided in improving completeness of data in our study. The outcome data in our study was collected by a researcher-in-residence from reading the documentation of HCPs (and was only 93% complete), hence in future, the recording of outcomes by the HCP's themselves might improve completeness of outcome data.
Field validation and compulsory fields within the app may have also contributed to the completeness of data. Saturations in oxygen for example, were not a compulsory field and were only recorded in 24 babies. This may reflect the power of compulsory fields but also a lack of time to wait for a second saturations reading once oxygen had been applied. Indeed, the adult pulse oximeters available took time to pick up a reading particularly on smaller premature infants. Local protocols (COIN [16]) specify how to measure oxygen saturations and that they should be taken in air and in oxygen as part of the assessment for starting CPAP (Continuous Positive Airways Pressure) but these were relatively newly implemented at the time of the study. Although completeness of data was high, there is always the possibility of false data being entered. Since we did not include any quality assurance in our study, we can only assume that HCPs were entering correct data.

Discussion of key fields within the NeoTree App
The percentage of low birth weight (LBW) admission <2.5kg may be a useful indicator for the procurement of feeding cups and NGTs, and the provision of Kangaroo Mother Care (KMC) beds.
These data could also potentially influence the maternal and obstetric department and potentially government policies to tackle the nutrition of Malawian mothers and their babies.
It is important to note that LBW or small for gestational age (SGA) is not the same as prematurity, hence why a maturity score was included within the NeoTree. The difference in maturity scores and estimated gestation exposed the inaccuracy of fundal height and length of pregnancy, suggesting significant underestimation of gestation using these methods. Feedback that maturity score was time consuming and required additional training prompted its removal from the NeoTree halfway through the study.
Our results from the subjective assessment of the severity of work of breathing (WOB) suggest that nasal flare and chest in-drawings were not considered 'severe' WOB. Head nodding, grunting and tracheal tug made up for the 9% of severe WOB. This could be further analysed in the next phase to improve understanding of the training needs of HCPs in assessing respiratory distress, and potentially develop a scoring system in the future.
Regarding examination of the newborn, the paucity of chest findings reported, and the complete lack of heart murmurs auscultated, suggest that related fields may not be appropriate for nursing cadres, but their relevance for doctors could be examined in the future. The flexible nature of the NeoTree app means these fields can be optional. For head circumference and birth weight, the app ideally could plot these automatically on a growth chart. However, it may represent a training challenge for HCPs to interpret these. Nevertheless, this is certainly a consideration for future iterations.

Limitations
Coverage was only 70%, so this may not be a representative sample. Despite the presence of paper charts, they could not be used as a reliable substitute for data as most charts have missing documentation [12], hence 100% coverage could not be achieved. However, since this was part of a proof of concept study, the digital form had to be completed additionally to the paper form, adding time and work-load to already pressured staff. The researcher-in-residence was present throughout the study, which may have enhanced uptake. There were difficulties completing antenatal fields, particularly when guardians accompanied the infant to the nursery whilst the mother was still recovering in the labour ward. As the study progressed, midwives started to bring the mother's labour ward notes, in addition to her hand-held record, with new admissions from the labour ward but the problem persisted for out-born babies. Hence the option of 'unknown' was added to many of the drop-down menus to preserve practical feasibility.
A major problem with our digital outcome audit is that the proposed system only collected data at the point of admission and discharge. What occurred during the crucial period between admission and death was not recorded, and therefore modifiable factors contributing to deaths and reciprocal solutions could not be identified. In turn, due to time constraints, steps five and six of the audit cycle could not be executed, leaving the audit loop unclosed. However, we have identified a considerable number of modifiable factors from patterns in aggregate data, hence, with the right resources and staff available, we could potentially close the audit loop.

Future Steps
In order to allow the scrutiny of individual causes of death, a free text field will be added to the 'Neo-discharge' form for the reviewing clinician to record (in a non-blame anonymous fashion) any possible modifiable factors that might have prevented that death. Copies of these 'death summaries' could potentially be printed and collated for review in monthly death audits, which would significantly increase the efficiency of these meetings and provide valuable contemporaneous insights into how and why an individual newborn had died. Other next steps include using the NeoTree where it completely replaces the paper form, or where no paper form exists in the first place, without the presence of the researcher on site. A study where clinicians or doctors use the app in another low-resource country would also be recommended.

Conclusions
Using an m-health application; the NeoTree, a digital outcome audit was successfully carried out by healthcare workers at a neonatal unit of a district hospital in Malawi with high completeness of data. These results were discussed at a local audit meeting and demonstrated that data collected digitally could stimulate quality improvement initiatives, such as improving thermoregulation of babies.
Limitations are noted in this study with only 70% coverage of all admissions. However, overall, this study illustrates how a digital audit using an app can improve documentation and richness of clinical data to help support delivery and configuration of local services. This study demonstrates huge potential for the use of a daily electronic record in low resource settings and these findings can inform the next stage of development for the NeoTree App, in particular, for guiding the development of linked data dashboards.

Figures
Example app screens.
HIV status of mothers of babies admitted to neonatal unit using the NeoTree app.