helpdesk utilisation

There were 253 309 MomConnect helpdesk messages recorded in DHIS 2 between 13 August 2014 and 31 March 2017. The final data set, after removing duplicates, had 241 715 unique messages, with 95 288 unique users. Almost half (48%) of the users sent more than one message. Messages from users that sent multiple messages made up 80% of all messages.

Figure 1 illustrates the mean number of messages received per day, by province, over time. The mean number of messages received per day was 252 (SD=162) messages, including Saturdays and Sundays. March 2015, September 2015, October 2015 and September 2016 had the highest number of helpdesk messages. After an initial rise, the number of messages stabilised, with a slight decline from November 2016 to February 2017. The low volume of helpdesk messages during November to January months contributes to the variation in volume of messages, but is reasonable, as this period corresponds with holiday months. Approximately 8% of all registered MomConnect users used the helpdesk feature. Helpdesk utilisation was highest from users in Gauteng (23.8%) and Kwazulu-Natal (20.5%) provinces and lowest from those in the Northern Cape (1.2%) (table 1). helpdesk utilisation over time and by province aligned with trends in the proportion of pregnant women registered on MomConnect and the proportion of pregnant women attending their first ANC clinic visit in the public sector (figure 2). For example, Gauteng also had the highest percentage of MomConnect registrations (22.1%) and ANC first visits (25.2%). Limpopo and North West provinces were exceptions. Limpopo had a relatively low share of both helpdesk users and helpdesk messages, despite having the third highest percentage of MomConnect registrations and ANC first visits. The proportion of helpdesk messages from North West Province was unusually high at 11.5%, but with only 7.6% and 6.6% of MomConnect registrations and ANC first visits, respectively. It is unclear why there was proportionately low helpdesk activity among MomConnect users in Limpopo and high helpdesk activity among MomConnect users in North West.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Mean number of incoming helpdesk messages per day, by month, by province.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Relative proportions of antenatal first visits, MomConnect (MC) registrations, helpdesk messages and registered helpdesk users, by province.

More generally, we found a surprisingly high number of helpdesk messages, given that the helpdesk feature is not specifically advertised. First-time registrants on the platform are invited to dial an unstructured supplementary service data (USSD) code to access other services, which include the ability to send messages to the helpdesk. Mothers are not informed about the helpdesk during registration and do not receive reminders about the availability of this service beyond the initial invitation. Comparably, in Zanzibar, a mobile phone intervention that provided MCH information via SMS and airtime vouchers for mothers (n=1311) to call health providers, found that 30% of mothers used the vouchers to call their providers for additional information in emergency and non-emergency situations.16 No other information was provided on the nature of the calls or effectiveness of the two-way communication component of the Zanzibar study. Participation in the two-way communication component of the intervention in Zanzibar was much higher than MomConnect registrants’ use of the helpdesk. However, it is important to note the larger scale of the MomConnect programme, the unadvertised nature of the MomConnect helpdesk and that vouchers are not provided to mothers.

Reasons for helpdesk use

The helpdesk interface (called CasePro) allows helpdesk staff to group the messages into eight categories: clinic code (to indicate a facility code needed to be assigned or corrected), complaints, compliments, message switch, opt out, prevention of mother-to-child transmission questions, ordinary questions and ‘Other’. Helpdesk messages in the DHIS 2 data set included the category variable. The majority of the messages (78.5%), were questions. The ‘Other’ category made up the second largest proportion (13.2%) and included spam, unable to assist, unclassified, duplicate and language switch messages. Table 2 provides details of the proportions of message categories by province. A χ² test revealed a significant association between message category and province (P<0.001).

Compliments and complaints comprised approximately 4% of the messages, with a higher relative probability of compliments from Gauteng than most other provinces. It is encouraging that there were more compliments than complaints submitted to the helpdesk; there were roughly four compliments to every complaint. The ratio of compliments to complaints in the DHIS 2 data set was slightly lower than previous analysis of data obtained from the helpdesk ticketing interface (called CasePro).15 As a whole, the small proportion of complaints and compliments shows that the primary value of the helpdesk to mothers is the ability to interactively seek maternal information that may not be included in the standard MomConnect message set.

helpdesk questions

For this analysis, messages that were categorised as a question and were in English (n=1 24 328) were further coded into 1 of 12 key topics. Key topics were systematically derived from the list of responses for FAQs and the stage-based maternal messages. The code list, including keywords for each topic, was finalised in consultation with the helpdesk manager. A key topic was identified in 65 364 helpdesk messages. These are summarised by province in table 3. It is important to note that misspellings, abbreviated language and slang that are often used when texting would have prevented keywords from being identified in the messages. We suspect this is the main reason why keywords were not identified in half of the English messages.

There was a significant association between key topics of messages and province (P<0.001). Questions coded as ‘baby’ (questions about the baby) and ‘breastfeeding’ were less likely to come from Limpopo, Northern Cape and Western Cape than Gauteng. HIV-related questions were more likely to come from KwaZulu-Natal and Mpumalanga than Gauteng, both by approximately 20%. ‘Support’ questions that indicated need for social or psychological support (eg, domestic violence, depression or seeking support groups) were 25% more likely to come from KwaZulu-Natal than Gauteng.

We suggest that the standard categories in the current helpdesk system be expanded to include more specific topics. This can allow for a more comprehensive understanding of issues that mothers are inquiring about. For example, we found that HIV-related queries were more likely from mothers that registered at facilities in KwaZulu-Natal and Mpumalanga. This corresponds with higher HIV prevalence in the two provinces,17 and also suggests the need to address knowledge gaps around HIV and pregnancy. The correlations of key topics and languages and provinces could help identify specific needs in different geographical areas of South Africa, and could allow for targeted or customised messages that can better address mothers’ language needs in specific locales.

helpdesk languages

Messages sent to the helpdesk could have been in any language used in South Africa. However, SMS responses back to mothers were only in English. Google Sheets’ ‘detect language’ function was used to identify the language of the incoming message and create the language variable. One limitation of the study is that the study team could not verify that the correct language was accurately detected for all messages.

The most common languages from the data set were English, Zulu, Xhosa, Sesotho and Afrikaans. Languages that represented <1% of the messages were categorised as ‘Other.’ Two-thirds (65%) of the messages were in English. Zulu was the next most frequent language at 17%, followed by Xhosa at 6%, Sesotho at 5% and Afrikaans at 2%. As one would expect, the language of incoming messages was significantly related to province (P<0.001). The correlation between the language and origin of the messages were internally validated; Zulu-speaking communities are common in KwaZulu-Natal and Mpumalanga, Afrikaans-speaking communities in the Northern Cape and Western Cape, Sesotho-speaking communities in Free State and Xhosa-speaking communities in Eastern Cape. See table 4 for details on the per cent of languages, by province. Notably, when comparing languages between KwaZulu-Natal and Gauteng, the relative probability that a message would be from KwaZulu-Natal was more than double when the language was in Zulu. However, the relative probability that a message would be from Gauteng increased five times when the message was in English.

Response to helpdesk messages

The average time for helpdesk staff to respond to all messages was 0.65 (SD=2.3) days, or approximately two-thirds of a day. However, this reflects that messages received on Fridays, Saturdays and Sundays were not responded to until Monday. The response times also varied by month, with longest response times during December 2015, and January 2016 and 2017, likely due to staff being on holiday during these months. There was a significant difference in response time between languages (P<0.001). Messages in Zulu and Xhosa had the longest mean response times, at 0.84 (SD=2.6) and 0.82 (SD=2.2) days, respectively. Messages in English and Sesotho had the shortest response times, at 0.61 (SD=2.2) and 0.62 (SD=2.3) days, respectively. A pairwise comparison of mean response times supported the significantly longer response time for messages in Zulu and Xhosa when compared with messages in English (P<0.001 for both Zulu and Xhosa), Sesotho (P<0.001 for both) and Afrikaans (P=0.028 for Zulu only).

Importantly, this study identified potential staffing gaps, such as the need for more Zulu-speaking and Xhosa-speaking staff given the comparatively long response times for Zulu and Xhosa messages. The majority of messages were in English and staff were able to quickly respond to English messages, compared with those in Zulu and Xhosa. Currently, only one helpdesk staff member speaks Xhosa. Additionally, keeping one to two helpdesk staff working during the holiday period could improve response rates during this time.

There exists early evidence of the helpdesk’s unique position to detect systemic flaws in the health system,15 which was possible because the helpdesk is positioned at the national office. This unique placement of the helpdesk is important for enabling the NDOH to strategically identify and address other national and provincial-level gaps in the health system. Additionally, the NDOH may also need to consider alternate strategies to effectively address changes at the facility level, since addressing issues at individual facilities likely requires a much higher level of effort from staff.