The IMCI strategy focuses on the most common childhood conditions that together result in over 80% of paediatric admissions and mortality in most African hospitals: malaria, anaemia, pneumonia, diarrhoea/dehydration, malnutrition, HIV/AIDS, meningitis and the sick young infant. These conditions are defined as IMCI diagnoses and all remaining conditions as non-IMCI diagnoses for the purposes of this report. The design of the PAR reflected both key IMCI symptoms and signs and approaches to severity classification and local information needs identified after meetings with hospital staff. The PAR was therefore administratively accepted as the first stage in the official hospital record and was not a 'study form'. For ease of completion the presence of many symptoms and signs could be indicated by circling options provided, although space was available for more traditional free text. Treatment and nursing observations were recorded on the usual hospital forms.

The Naivasha District Hospital was selected in collaboration with the Division of Child Health in the Ministry of Health for this pilot work as it was reasonably accessible to the study team (based in Nairobi) and provided an anticipated paediatric workload of approximately 1800 admissions per annum. Permission from the hospital management team was sought before embarking on the study which was explained to staff at a subsequent open meeting. We conducted a three-hour training workshop for clinicians and nurses to introduce the PAR. Lectures, discussions and videos were used to explain and demonstrate the symptoms and signs. Attendance was not compulsory and no incentives were offered.

Thirty-two members of staff including 2 medical officers, 6 clinical officers (health workers with 3 years medical training), 18 nurses and 6 others (medical records, X-ray, laboratory and pharmacy staff) were trained during the initial workshop. This initial training included 6/16 of the clinicians who might be responsible for admitting children. Amongst these six were most of those assigned to routine, daytime paediatric services at the time the PAR was introduced. We conducted five more training sessions on the PAR over the study period in order to include all clinical staff involved in the care of children. The total number of staff trained in use of the PAR at the end of the study was 67 including 15 of the 16 clinical officers in the hospital. All training was conducted within the hospital.

At the hospitals request, we also provided 9 hours of training on paediatric inpatient management based around WHO guidelines 7. At the start of the study only one clinician had received IMCI training, no further staff received IMCI training during the intervention period. Clinical officers responsible for admitting children were each assigned a unique identifying code known only to the research staff at the start of the study.

The study period was divided into pre and post intervention by the PAR introduction. The first PAR training was conducted on 17^th May 2004 and the PAR was introduced on 24^th May 2004 to replace the existing paediatric medical records. Thus the period before the 17^th May 2004 was regarded as the pre-intervention period and from this period we retrospectively selected 168 consecutive paediatric case records over a two months period (from 3^rd March 2004 to 3^rd May 2004). These records were scrutinized for documentation of symptoms and signs specified in the PAR and data abstracted. We then randomly, prospectively selected 60% of all calendar dates during an 8-month study period aiming for approximately 150 admission events per 2 months and abstracted the same data from the records of children admitted on those dates. This 8 months period is referred to as the post-intervention period. During the post-intervention period feedback on the use of the PAR was provided at an open meeting held every two months where staff had a chance to ask questions and give suggestions. These feedback meetings therefore divided the post-intervention period into four blocks of two months.

In the post-intervention period only data present on the PAR was abstracted. Before intervention the respiratory rate and a history of fever were rarely documented in the clinical record. However, on a number of occasions a measured temperature and a record of respiratory rate were documented by nurses at admission perhaps obviating the clinician's need to include these in the clinical record. We therefore did not assess the completeness of documentation in the pre-intervention period for these specific signs nor for the assessment of conscious level using the AVPU scale (Alert, responds to Voice, responds to Pain, Unresponsive) that was not in use before intervention.

Data were entered on site by the investigators using an entry programme with in-built range and consistency checks designed to minimize data entry errors. Personal identifiers were not recorded in this database.

Data were cleaned and analysed using STATA v8.0 (Stata Corporation, USA). The documentation of IMCI emphasized symptoms and signs over time is described both in all records and for specific, common diagnoses. The χ² test and Odds ratio are used to examine associations between use of the PAR and diagnostic group (IMCI or non-IMCI). The χ² test for trend and odds ratios are used to examine for an association in the use of the PAR with calendar time. To explore whether the record of the presence or absence of clinical signs on the PAR had clinical significance rather than being a hurried and inaccurate or random record of the clinical consultation we examined whether signs supposed to indicate severe disease were associated with increased mortality. To do this we used computer algorithms based on IMCI symptoms and signs for two common diseases, malaria and pneumonia, to define severe disease based solely on the recorded clinical data. We examined the association of both these derived severity classifications and the clinician's overall diagnosis of the severity of disease with death as the gold standard indicator of severe disease. For the purposes of these analyses pneumonia classifications were dichotomised as very severe or not and malaria as severe or not. This approach provides some indication of the internal validity of the data recorded on the PAR.

Ethical approval was obtained from the KEMRI National Ethics Committee and the Kenyan Ministry of Health.

Before intervention, the number of newborns less than 7 days old (not covered by the current IMCI strategy) was very small (Table 1). During this time most sick babies born in hospital were considered part of the mother's admission and had no separate inpatient number or medical record. Throughout the period of data collection consistently more than 80% of the children admitted had an IMCI diagnosis. Overall mortality was 13% (Table 1).

Sixty nine percent of the children admitted during the post-intervention study period had a PAR used. The use of the PAR increased over time from 50% (post-intervention period 1) to 84% (post-intervention period 4) at the end of the study [score test for trend, p < 0.001, odds ratio per unit increase in time period, 1.37, 95% CI 1.14–1.63), p < 0.001 (table 2)]. There was considerable variation between clinicians in the use of the PAR ranging from 53% to 100% (Figure 1) and this was unsurprisingly associated with the date of their training (data not shown). Use of the PAR was also found to vary depending on the diagnosis. Forty six percent of children with a non-IMCI diagnosis had a PAR used whereas 73% of children with an IMCI diagnosis had the PAR used [OR 0.32, 95% CI 0.21–0.41, p < 0.001 (Table 2)].

There was a marked improvement in documentation of important general clinical features like pallor and level of consciousness for the admitted children. There was less impressive improvement in documentation of others, notably weight and immunization status (from 55 to 58% and 1–21% at the end of the study respectively, table 3). Before the pilot study began signs known to be important in classification of severity of common, specific illnesses 7 were rarely recorded (if at all). During the study we saw a marked improvement in documentation of signs for classifying the severity of malaria, pneumonia and dehydration (table 3). For example the proportion of children admitted with a diagnosis of pneumonia who had a record of the key sign of indrawing increased from 33% in the pre-intervention period to 91% in the final post-intervention period. The fact that a diagnosis specific analysis (that is based on all records, not just those when a PAR was used) shows greater improvement in documentation than the overall analysis suggests that documentation improves if the sign is perceived as directly relevant to the diagnosis.

To examine whether the improvement in the quality of documentation was simply a result of the presence of the intervention team (a Hawthorne effect) we analysed the quality of documentation in relation to PAR use in the post-intervention period only. We found that when the PAR was used documentation for signs relevant to specific diagnoses was almost always more than 90% for key signs compared to <50% when clinicians still used traditional free text notes (table 4). The analyses exploring the internal validity of the data recorded on the PAR (whether the data had clinical meaning) showed similar associations with death for the clinician's classification and the computer derived classification that was based on the recorded signs (table 5). This suggests that clinicians were not just mechanically 'ticking the boxes' when completing the PAR to oblige the study team but that the recorded data did have clinical meaning.

In conclusion, we have found that the standardized paediatric admission record was well accepted by clinicians and it greatly improved documentation of illness for most admitted children. This form may be a useful tool in the implementation of the Division of Child Health guidelines for care and IMCI at district hospitals. Future work should consider the possibility of incorporating both clinical and nursing records into a unified format that may further enhance the quality of records and promote efficient use of time.

CI – Confidence Interval

IMCI – Integrated Managemnt of Childhood Illness

OR – Odds Ratio

PAR – Paediatric admission record

WHO – World Health Organisation

The author(s) declare that they have no competing interests.

Dr. Sekela Mwakyusa was responsible for undertaking the study, collecting the data, data management, analysis and writing the manuscript.

Dr. Annah Wamae, Prof. Aggrey Wasunna, Dr. Fred Were, Prof. Fabian Esamai, Dr. Bernhards Ogutu, Dr. Assumpta Muriithi, and Dr. Norbert Peshu contributed to the design of the study, oversaw its implementation, reviewed the analysis and contributed to drafting of the paper.

Dr. Mike English designed the study, obtained funding and supervised data collection, data management, analysis and the writing of the manuscript.