This study demonstrates that providing vaccine-related targeted education to mothers at home is an effective and practical strategy to improve childhood immunization rates in low literacy settings such as ours. In this randomized controlled trial, a significant improvement in infant DPT-3/Hepatitis B vaccine immunization rates was observed in the group of mothers who received home-based education on the importance of vaccines, compared to those who received standard health promotion messages only.
In low-income countries, efforts to improve infant vaccination completion rates have focused primarily on supply and/or provider factors, with little focus on creating demand for infant immunization services. The major thrust of the supply-side interventions to improve vaccination rates has been through mass immunization campaigns. These campaigns have been successful in improving vaccine coverage rates . However, there are certain drawbacks of mass campaigns such as those done for polio vaccine in Pakistan and other developing countries [24–26]. They lead to a misconception on the parents' part that the child will be delivered all vaccines at home , and can result in a decline in the number of visits to the immunization centers, paradoxically driving down routine immunization coverage rates. Furthermore, these mass immunization campaigns have resulted in declining performance of routine EPI activities in Pakistan.
Educational interventions have been successful in raising awareness regarding vaccine and increasing demand. Jacobson et al  were successful in increasing pneumococcal vaccine coverage rates among the elderly, by using low-literacy pamphlets encouraging study participants to "ask your doctor about the pneumonia shot". Kimura et al  were able to increase influenza vaccine coverage among workers in long-term care facilities with the help of an educational campaign and provision of free vaccines. In Pakistan, Usman et al  reported an increase of 31% in DPT3 completion among infants of mothers who received primary healthcare center-based education on their first immunization visit.
The success of educational interventions in modifying health-seeking behavior may also be attributed to the focused nature of the interventions. This is certainly true in our study. The intervention group received a 5-minute educational session, focusing on the importance of immunization for a child's health. The control group, on the other hand, received a 10-15 minute verbal session on general child health promotion. Therefore, the group receiving the focused message may have been more likely to understand and retain its content, and modify their behavior, compared to the control group who may have had "information overload".
We observed that maternal knowledge/perception regarding importance of vaccines was significantly associated with higher DPT-3/Hepatitis B immunization rates (RR = 2.11; 95% CI: 1.33 - 3.34). Our results are consistent with findings of other studies [4, 9–11]. Surprisingly, infants living in rented houses (considered a crude proxy for lower socioeconomic status) were more likely to have received all 3 doses of DPT-3/Hepatitis B vaccines as compared to infants living in houses owned by their families (RR = 1.26; 95% CI: 0.93 - 1.71). However, this association is not significant. Another explanation could be that many of the households in these populations are considered illegal squatters but claim ownership of the land on which they've built their house. Therefore, paradoxically, households in rental accommodation could actually be socio-economically better-off.
It is worth noting that even with the educational intervention specific to vaccines, only 72% of infants were fully immunized in the intervention group. This is because our study included very low-income, low-literacy populations of Sindh province where baseline immunization rates are 48% , much lower than the national reported figure of 73% . The national average is a composite figure, including more prosperous and literate parts of the Pakistani population and the large province of Punjab which has a more functional EPI system and estimated vaccine coverage of 65% . National surveys may overestimate or over-report vaccine coverage rates . Different immunization centers use different vaccination cards, which differ in the design and method of recording proof of vaccination. This makes it difficult for data collectors to accurately determine an infant's immunization status. Furthermore, verbal reports, in lieu of vaccination cards, are often accepted as proof of immunization, but our experience using serological confirmation shows poor correlation between verbal recall and serological immunity . Therefore, the true vaccine coverage rate for Pakistani children may be closer to the figure of 59% estimated by the recent Demographic and Health Survey of Pakistan . Although ascertaining the reasons for low vaccine coverage was beyond the scope of this study, many barriers to improving immunization coverage remain in low-income communities and need to be systematically addressed.
Our study also has a few limitations. First, the third key message provided education on retaining vaccination cards. Therefore, mothers in the intervention group were 17% more likely to save the card and provide proof of vaccination for outcome assessment. However, as shown above, using the stringent proof-of-vaccination via card to determine outcome did not bias our results significantly if infants with maternal recall of vaccine receipt were also included as fully immunized. Including these infants as fully immunized, DPT-3/Hepatitis B immunization rates in the intervention group were still 18 percentage points higher than the control group.
A second limitation is the lack of blinding of CHWs and participants as the intervention was educational in nature. However, the investigator (BH) assessing the outcome four months after the intervention was administered was blinded to the exposure status of the study participants. Furthermore, chances of spillover effect, or contamination between the intervention and control arms were minimized by choosing mother-infant pairs from five different communities in Karachi, lowering the probability that households participating in our study with an eligible newborn would be located close to each other. We also observed a trend for mothers in the intervention group to be more educated compared to those in the control group (34% vs. 25%). However, this difference was not statistically significant and other measures of parental knowledge about vaccines did not favor the control group. Our study also had a high refusal rate (27%) which may have excluded participants less likely to accept vaccines from the trial. However, the most common stated reason for refusal was absence of the child's father when study staff visited the household for initial recruitment.
Our educational intervention has the potential to be cost-effective. The cost of the intervention per CHW was estimated to be Pakistan Rs. 80 ($1). This includes the cost of laminated colored pictorial cards used by the CHWs to educate the mothers in the intervention group, as well as pamphlets of the pictorial messages left at each participant's house. We also estimated the cost of scaling-up this intervention nationally, through the Lady Health Worker Program. Given that there are 100,000 lady health workers working all over Pakistan, we estimate that the cost of the national scale-up will be approximately $200,000 for the national program ($100,000 for the cards and pamphlets, and $100,000 for training sessions).