Factors affecting medical file documentation during telephone triage at an emergency call centre: a cross-sectional study of out-of-hours home visits by general practitioners in France

This was a cross-sectional study of patients calling the medical emergency call centre of Créteil (Henri-Mondor hospital, district of Val-de-Marne, Paris area) between January and December 2015 and for whom an OOH-HV was prescribed.

The endpoints were the absence of the key indicators in the medical file: medical interrogation, suspected diagnoses, and ruled-out severity.

Data extracted from the medical file were the day and time of the call (weekday, weekend, public holiday, holiday, 00:00–04:00 am, 04:00–08:00 am, 08.00–00.00 am) patient characteristics (age (< 15, 15–64, > 65 years), sex, RFE (see Table 1), medical interrogation, diagnostic hypothesis, ruled-out criteria for severity (i.e., if at least one negative sign was reported, such as “no chest pain”, “no dyspnea”, etc., severity criteria was considered reported), diagnostic code and decision after a face-to-face encounter (remain at home, referred to the hospital).

We performed a descriptive analysis of the RFE (by age class, date and time of call) and key indicators in the electronic medical file. Categorical data are reported as number (%) with 95% confidence intervals (CIs) and continuous data as mean ± SD or median (interquartile range [IQR]). After testing for linearity gap, continuous data were categorized for analysis. We used three univariate analysis rounds, one for each endpoint (unreported medical interrogation, diagnostic hypothesis, ruled-out severity criteria), by using logistic regression.

After applying the RAM to achieve consensus on the main strata, to be sampled, the expert panel identified flu epidemic, bronchiolitis, heat wave, pollution peak, and holidays and public holidays, for 10 determinants (or factors) of stratification. Each strata could contain several determinants (“heat wave” occurring during summer “holidays”) (see Additional file 1).

Among the 10,284 OOH-HVs performed from January 1, 2015 to December 31, 2015, we selected for analysis 748 unique electronic medical files with an OOH-HV performed during the randomly selected 30 days (19 holiday or public holiday days, 2 weekend days and 9 weekdays of worked weeks) (Fig. 1). For 708 records (95%), medical triage was performed by GPs and for the other 5% by EPs.

In all, 455 callers were women (61%) and the median age was 33 [IQR 8–58]. When patients were children (age < 15), the median age was 2 [1‐6]. Callers were largely bystanders (58%, n = 436) or patients themselves (31%, n = 231). Overall, 6% (n = 48) of the calls were from emergency first responders. Calls were predominantly received before midnight (61%, n = 455).

RFEs were digestive tract symptoms (22%, n = 163); fever (19%, n = 141); ear, nose and throat symptoms; cardiovascular, respiratory and dorsolumbar problems (6% each); and dizziness and fainting (5%, n = 41) (Table 1). RFEs were not reported in 29% of files (n = 216). After the OOH-HV, 8% of patients (n = 58) were referred to a hospital.

Medical interrogation was not reported in 16/748 files (2% [95% CI 1.2–3.4]). The diagnostic hypothesis formulated by the triage GP was not reported in 432/748 files (58% [54–61]), and ruled-out severity criteria were not reported in 449/748 (40%) (Fig. 1). Only 14/748 files (2% [1.0–3.1]) had every quality criterion present, and 9/748 (1% [0.5–2.3]) had none of the criteria.

For unreported medical interrogation, factors associated with altered quality of reporting were a call received during public holidays (reference weekday, OR 14.7 [2.9–73.2]), duration of call by triage assistants (reference ≤48 s, OR 1.1 [1.1–1.1]) and triage physicians’ telephone occupation rate (reference ≤33.7%, OR 0.5 [0.3–0.9]) (all p < 0.02) (Table 2). For unreported diagnostic hypothesis, factors were patient age < 15 years (OR 1.5 [1.0–2.1]), digestive RFE (reference general, OR 1.5 [1.1–2.3]), number of incoming calls to triage physicians (reference ≤9700, OR 0.8 [0.7–0.9]), call duration (triage assistants, reference ≤48 s, OR 1.1 [1.1–1.1]; physicians, reference ≤147 s, OR 1.2 [1.1–1.3]) and physicians’ telephone occupation rate (reference ≤33.7%, OR 0.8 [0.7–0.9]) (p < 0.05). For unreported ruled-out severity criteria, factors were the hour of call (reference 8:00–00:00 am, for 00:00–04:00 am, OR 0.6 [0.4–0.8]), weekend call (reference, weekday call, OR 0.5 [0.3–0.9]), workforce at the call centre (assistants, reference 3, OR 1.2 [1.1–1.4]), number of incoming calls (assistants, reference ≤30,000, OR 1.2 [1.1–1.3]; physicians, reference ≤9700, OR 1.2 [1.1–1.4]), assistants’ call duration (reference ≤47 s, OR 1.1 [1.1–1.1]) and telephone occupation rate (assistants, reference ≤26.2%, OR 1.3 [1.1–1.5]).

On multivariate analysis, unreported interrogation remained linked to assistants’ call duration (reference ≤47 s, OR 12.9 [2.3–72.5]), unreported diagnostic hypothesis to assistant’s call duration (OR 2.1 [1.2–3.7]), and unreported ruled-out severity criteria to assistant’s call duration (OR 1.8 [1.1–3.1]) and telephone occupation rate (reference ≤26.1, OR 1.8 [1.3–2.5]) (Table 3).

Only 2% of the medical files for patients receiving an OOH-HV after a call at a French urban emergency call centre included all analysed quality criteria. On univariate analysis, individual data, such as age, sex, or the RFE, were not responsible for poor reporting in files. Factors associated with incomplete reporting appeared to be organizational only (workforce, workload, time of call, telephone occupation rate) and on multivariate analysis, remained significant for non-medical triage assistants only.

Regarding the RFE, our results were globally similar to other European settings [13]. Organizational factors, such as workload, are known to be negatively associated with the quality of a given procedure [14, 15] and should be anticipated to enhance the reliability of a given system. Our study suggests that only triage assistant-linked factors affect the medical reporting. Prolonged interrogation by an assistant may be due to initial difficulties (unstructured interrogation and/or lack of communication skills [16]) that are discussed with the GP during the call transfer. Also, the longer the assistant’s interrogation, the more information that may accumulate [17]. The GP may benefit from unrecorded information. This situation could explain why the GP’s workload and call occupation did not affect our endpoints: if the assistant presents an interrogation as particularly difficult, only the most important components may be recorded (severity, diagnostic hypothesis). Furthermore, as compared with non-physician staff, and possibly due to differences in communication structure and decision process, GPs may need shorter communication time in telephone triage to prescribe an OOH-HV [18].

Two Dutch studies reported poor results for quality of telephone triage [19, 20]: although clinical problems were quasi-constantly reported, less than 30% of mandatory questions were asked, personal situation and medical history were missing in more than half of the files, emergency was underestimated in 41% of the calls, data on home management and safety-net advice were reported in 40% of patient files, clinical evaluation was imperfect, and triage outcome was appropriate for only 58% of the patients. Moreover, some unasked mandatory questions were still reported. Nonetheless, one of these studies focused on triage assistants, and calls were made by incognito-trained laypersons. Similar to our findings, individual data, including RFE, were not associated with altered results, which suggests that the system is equitable.

French data on public telephone triage are scarce and mostly focused on the impact of a single event on emergency call centre activity [21]. A recent monocentric study analysed the key performance indicators of an emergency call centre. The most important factor affecting answer time (< 20 s) was the overall telephone occupation rate [22]. This study showed significant temporal trends for the number of incoming calls, especially between 20:00 and 00:00 am. Our results showed the same evening trend in number of calls (61% incoming before midnight).

In Europe, telephone triage is not standardized [23]. Systematic physician-led triage is not supported by high-level evidence and has not been evaluated in terms of quality of reporting in the patient’s medical file but rather safety and efficacy [24‐26]. GP-led triage, despite differences in decision-making process and information gathering [27] and longer training and experience as compared with nurses, showed similar safety and efficacy results. However, GPs could be more efficient in complex triage situations [28]. An optimal triage team could involve both staff types (nurses and GPs) [29]. The quality and safety of telephone triage remains an ongoing issue needing to be evaluated in methodologically robust prospective studies [30].

This is the first study to evaluate the typology of OOH-HVs by French GPs and the quality of medical files after telephone triage in an emergency call centre. In addition, it is the first time in France that the quality of information reporting in the medical files has been analysed with individual patient data. This study describes and analyses a wide spectrum of the overall process leading to an OOH-HV after telephone triage and helps understand the factors affecting the quality of reporting in the medical file.

One of the major limitations of our study is its monocentric character, which limits generalizability. Yet, because the electronic patient administration file’s software is the same in the four largest emergency call centres of the Paris area, the reporting in other centres might be similarly affected by factors linked to the system’s ergonomy. Moreover, and because many triage GPs share their activity between several call centres in the Paris area, there might not be a significant centre effect in these results.

Although our study suggests that patient data did not affect the quality of reporting in medical files after telephone triage, we cannot conclude formally because of the low number of variables tested, due to data-mining limitations. As well, we did not take into account the over-sampling fraction of weekends, which may imply selection bias in our adjusted regression analyses.

Current practices do not take into account patients referred to the hospital on their own but only those using ambulances, so the overall number of hospitalized patients after an OOH-HV (8%) may be underestimated. However, in many of the analysed files, the low referral rate suggests that the OOH-HV was the most appropriate decision after triage.

In addition, the quality of reporting may not be linked to the quality of interrogation. Why the quality of reporting remained linked to assistants-related work factors only is unclear. In the absence of a qualitative evaluation of the overall triage process, including communications between assistants and physicians, some factors may remain hidden.

The choice of key indicators was based on the French recommendations for the triage process. Whether their type or number was sufficient is unclear. Some other composite key indicators may be more exhaustive and may lead to other results [31]. For example, we chose to analyse the criterion “suspected diagnosis”, because we believe that it is part of a relevant triage process: when a triage GP formulates and reports one or several diagnostic hypotheses in the medical file, one can reckon that the final decision is adapted to the patient’s need. This choice is discussable in our setting, as the clinical assessment should be left to the mobile GP performing the OOH-HV: the main goal of telephone triage is not to diagnose a given pathology, but to assess the need of the patient. Because of a poor assisted coding process in the administration system at the call centre (ICD-10), we performed systematic data mining and recoded the RFE with the ICPC-2 classification. This could have biased our descriptive results but did not alter the endpoint “diagnostic hypothesis”. Moreover, because our study did not focus on the quality of triage itself, the endpoint “unreported severity” was considered negative when a single criterion was reported. Although this coding was potentially incomplete (“no purpura” present but “no meningeal syndrome” absent in febrile headache), the triage GP may have considered ruling out the severity by orally asking questions that eventually were not reported [20]. Nevertheless, even in this case, GPs may not rule out every severity criterion, because mandatory questions were regularly omitted in other studies.

First, because in our study, quality in medical reporting was affected by the number of incoming calls and the number of triage assistants and their telephone occupation rate, research in the French setting should first identify factors predicting increases in emergency call centre workloads and lack of staff [21, 32] and model them according to relevant variables, then evaluate the efficiency of corrective measures (recall of triage assistants or physicians). Second, it should identify individual factors associated with altered reporting in terms of triage staff, to identify the settings associated with lower quality of triage and reporting in medical files. Qualitative studies should then be conducted to analyse human factors affecting the process of call handling, medical interrogation and oral transmission between assistants and physicians. Third, because these factors could be educational, research should evaluate the impact of professional education for emergency call centre staff.

Practically, this study could help establish corrective measures in the emergency call centre. First, these measures could consist of targeted increases in the workforce, based on peak workload, especially during public holidays. Second, after feedback from our results, staff in the emergency call centre could undergo practical training in medical-file reporting. Quality criteria need to be better reported, and both the patient’s condition and final decision should be systematically assessed after the OOH-HV. This process is crucial because employees need to be aware of the importance of good traceability, for quality of the healthcare chain (transmission of data to other health stakeholders, or in case of change in assistant or triage staff) or the legal implications of poor reporting.