Using epidemiology in practice

Field epidemiology follows a recognisable sequence, and knowing its order prevents the common mistake of leaping to a cause. First, establish that an outbreak exists: compare the current count with the expected number for that place and season, and with the previous weeks and years. Second, verify the diagnosis, because a laboratory error or a change in testing can create an apparent epidemic out of nothing. Third, write a case definition with three parts — clinical features, person, and time and place — and use it unchanged while the investigation lasts. Fourth, and most productive, describe the cases by time, place and person. Fifth, generate a hypothesis about the source and the mode of transmission from that description. Sixth, test the hypothesis, usually by comparing the ill with the well: a cohort study calculates an attack rate for each exposure, while a case-control study compares what the ill and a comparable group of the well were exposed to. Seventh, carry out environmental and laboratory investigations where they are useful. Eighth, institute control measures, which need not wait for the end of the investigation. Ninth, communicate the findings to the community, to the authorities and to clinical staff. Tenth, maintain surveillance after control to confirm that the outbreak has truly ended. Throughout, the team writes as it goes, because an investigation that is not recorded cannot teach anyone afterwards, and the written record is what allows a later review to ask whether the control measures worked.

Describing cases by person, place and time is the analytical heart of an outbreak investigation, and each axis carries its own information. Time is the most powerful. Counting the onset of illness and drawing the cases as bars, one per day or per hour, produces the epidemic curve, whose shape is itself evidence. A curve that rises steeply to a single peak and falls within roughly one incubation period indicates a point-source exposure, in which many people were exposed during one short event such as a single meal. A curve that rises gradually and shows successive peaks separated by about one incubation period suggests person-to-person spread or a continuing common source, because each generation of cases produces the next. Place is drawn as a spot map of where cases live, work, eat or draw water; clustering points to a shared vehicle such as one well or one food handler. Person is explored by comparing attack rates between groups, for example children under five against older children, or those who attended a gathering against those who did not. Attack rate, not the raw number of cases, is the correct comparison, because groups are not the same size: forty cases in one village and twenty in another may mean the smaller village was at twice the risk. The same descriptive discipline is used outside emergencies. Routine surveillance data limit themselves to person, place and time, and the epidemiological bulletin that results is what allows a health team to see a district as a whole. Prevalence answers how many people have the problem now; incidence answers how many are becoming ill; the case fatality rate answers how lethal it is among those affected. Together these measurements direct scarce resources to the place and group where the risk is highest.

Every epidemiological statement inherits the quality of the data beneath it, and much of that data is created by nurses and community health workers in registers and tally sheets. Three properties decide whether a dataset can be trusted. Completeness asks whether every case was recorded; a disease that is not reported cannot be counted, and under-reporting makes a programme look better than it is. Consistency asks whether every reporting point uses the same case definition and the same period, because a clinic that counts suspected and confirmed cases together will always appear to have more disease. Timeliness asks whether the report arrives early enough to allow action; weekly surveillance that is submitted monthly has lost its purpose. Systematic biases deserve their own names. Ascertainment bias arises when a disease is more thoroughly looked for in one group than another, for example in the village nearest the clinic. Denominator error arises when cases are counted but the population they came from is unknown or wrongly estimated, which distorts every rate calculated from them. Recall bias arises when people remember an exposure differently because they are ill, which is why interviewers ask the same questions of the ill and the well. The nurse stands at three points in this system. She is a case-finder, applying the case definition to the patient in front of her and checking that stool, blood or sputum specimens are taken and that specimen containers are sealed and labelled with the date and place, because a mislabelled or degraded sample can make an outbreak look like nothing at all. She is a recorder, entering the date of onset, the age and the village rather than a vague note. She is also a communicator and teacher: she tells the community why the questions are being asked, which improves the recall and honesty of the answers, and she reports notifiable diseases promptly, because a single silent case is enough to seed the next generation of cases. WHO guidance on surveillance and on vaccination coverage places the same emphasis on these human steps, since no statistical method can repair data that were never collected, were collected twice, or were collected differently in every village.