To estimate the drug use indicators accurately, it is important to follow specific procedures for drawing samples of health facilities and patient encounters. These procedures will vary depending on the objectives of the study and the availability of data. The recommended process for sampling facilities and encounters in the basic drug use survey is described below.

Drawing a sample of health facilities

In order to draw conclusions about drug use practices in an area with many health facilities, it is necessary to select a representative sample of these facilities (usually 20). Drawing this sample in a haphazard way can bias the results. This section provides methods for drawing a systematic random sample of facilities in a way that will ensure maximum confidence in the estimates. The following steps should be followed.

Step 1: Identify the sample area and type of facilities for the study

First identify the geographic areas in which the survey is to be done. If there are to be study and control areas, develop and record clear criteria for including facilities in either group.

Choose the types of health facilities that are to be included in the study, for example, hospitals, health centers or dispensaries. Make a list of the names and locations of all these facilities; the sample will be taken from this list.

Step 2 (OPTIONAL): Organize the areas and facilities into groups

The health facilities to be included in the study can often be organized into groups. Sometimes the facilities are naturally grouped by geographic location, perhaps by district or by urban-rural location. Another natural grouping factor might be level of service, for example, multi-physician polyclinics versus dispensaries with only paramedical staff. One way of increasing the precision of indicator estimates is to ensure that these groups are appropriately represented when the sample is drawn, rather than running the risk that one group is over- or under-represented.

The easiest way to ensure an appropriate balance between these groups is to organize the list of facility names created in the previous step into groups before taking the sample. For example, put urban facilities at the top of the list, followed by the rural facilities. It is also possible to organize the list by more than one characteristic. For example, if three separate administrative districts are to be studied, group the facilities first by district. Within each district, list the urban facilities, then the rural ones. This grouping and pre-sorting will help to ensure that the sampling procedure will allocate an appropriate number of health facilities of each type.

Step 3: Select facilities by systematic random sampling

To draw a systematic sample, begin by numbering all facilities on the list. Then calculate a sampling interval by dividing the total number of facilities on the list by the number of facilities to be included in the sample (usually 20). For example, if there are 53 health centers in the study area and 20 are to be visited, the sampling interval is 2.65.

Now you have to choose a point on the sampling list from where to start your sampling. Choose the first facility in the sample as follows. Round the sampling interval up to the highest full number (in this case 3). Choose a random number between 1 and this number (in this case 1, 2 or 3). This can be done by using a table of random numbers, the latest figure of the number on a banknote, or simply by dice or paper lots. There are also electronic calculators that can select random numbers.

To identify the next facility to include in the sample, add the sampling interval to the previous result, and round up to find the facility number to include. In the example, if the first sample was facility 2 from the sampling list, the next three facilities selected would be number 5 (2 + 2.65 = 4.65, rounded up to 5), number 8 (4.65 + 2.65 = 7.30, rounded up to 8) and number 10 (7.30 + 2.65 = 9.95, rounded up to 10). Continue with this process until all facilities have been selected.

When the objective of the study is to compare indicators for two groups, the list should be divided and each subgroup should be treated as a separate list, with a separate equal-sized systematic sample drawn for each subgroup. Drawing an equal number of facilities in each subgroup ensures that the comparison of the subgroups will be as accurate as possible.

Step 4: Keep a record of the sampling frame

The list with the groupings from which the sample was selected is known as the sampling frame. This is the record of the sampling process, and can also be used to draw additional samples during future activities. If separate samples were drawn from this list to compare two groups, the sampling interval for each group should be recorded on the list.

Drawing a retrospective sample of patient encounters

The basic indicator study calls for a sample of 30 prescribing encounters per health facility, or 100 encounters if prescribing practices in individual facilities are to be compared. The following section describes how to draw these encounters from historical patient records.

Step 1: Confirm the availability and accessibility of medical records

Prescribing records can be organized in many different ways. Before actually implementing a retrospective sample it is necessary to become familiar with the organization of medical records in a few facilities, to ensure that it is possible to extract data from them in an efficient and reliable way. If the data systems in the intended sample of facilities are all similar, the accessibility of data needs to be tested at only a few sites. If there are substantial differences between facilities (based on size, urban-rural location, level of staffing, etc) the record systems and procedures should be tested in a few sites in each subgroup to be certain that the proposed sampling methods will work equally well in all types of facility.

Possible sources of retrospective prescribing encounter data include clinic registers, health worker treatment logbooks, patient- or family-files, or some type of pharmacy record (such as retained prescription forms). Particular questions to be answered at this preparatory stage are:

• What is the source of the chronological listing of patient encounters from which the sample will be drawn?

• What is the source of data on patients, providers, health problems, and drug treatments?

• How are the records stored, and are they available for the intended study period?

• If two or more record sources must be linked to collect information from one encounter (e.g. patient records and prescriptions kept in the pharmacy) what is the success rate of making this linkage?

If it is found during these preliminary tests that historical data sources are more incomplete than expected, or that the necessary medical record data are too difficult or time-consuming to extract, it may be preferable to switch to prospective sampling. However, if the record sources appear adequate, the process for sampling historical encounters, as described below, can be used.

Step 2: Locate encounters to be included in the sampling frame

Locate the chronological listings of all patient visits made during the selected study period, e.g. the twelve-month period prior to the survey date. These listings will constitute the sampling frame of prescribing encounters. If not all records for this period can be found, the list should cover as much of the study period as possible.

Potential biases to be avoided include: excluding or under-representing one or more providers in a facility because of missing records or misplaced log books; under- or over-representing some types of disease and treatments due to missing data from certain seasons during the study period; or excluding encounter records of a certain type, for example, prescriptions filled outside the health facility.

Put the available listings in a systematic order, by date if they are assembled that way, or perhaps by provider, if individual patient logs are kept. Record the dates for the study period covered by the available listings on the facility summary form for each facility.

Step 3: Select encounters at regular intervals over the study period

If the patient encounter listings are ordered by date, it is easy to spread the sample encounters evenly over the period represented. The total number of encounters to be sampled in each facility has already been determined by the purposes of the study (30 for a cross-sectional study, 100 to compare facilities). In a process similar to the one used for selecting facilities, dividing the number of days represented in the sample frame by the number of encounters to be selected gives a sampling interval.

For example, if there are 365 days (one year) in the chronological sample frame and 30 encounters to be drawn per facility, the sampling interval is 12.2 calendar days (about one encounter every 12 days). Start the sampling process at the first day represented in the chronological sample frame. Select an encounter from the listing for this day (procedure described below). Subsequent encounters should be selected by skipping the appropriate number of days indicated by the sampling interval. For example, when the sampling interval is 12.2 the second case is selected from the 14th day of the listing (1 + 12.2 = 13.2, rounded up to 14), the third from the 26th day (13.2 + 12.2 = 25.4, rounded up to 26), and so forth.

From each selected day a single encounter should be picked at random by multiplying the total number of encounters listed for the day by a random number between 0.0 and 1.0, and rounding upwards. If no source of random numbers is readily available, follow a procedure to spread the encounters selected over different times of the clinic day. For example, for the first selected day pick the encounter from the beginning of the patients listed for that day, for the second selected day from the middle, and so on.

An alternative method can be used when the total number of encounters over the study period is known or can easily be estimated. In that case the logbook can be used as a sampling list. Divide the total number of encounters in the study period by the desired size of the sample to obtain the sampling interval. For example, if about 5000 encounters were recorded in the study period and a sample of 100 is needed, the sampling interval would be 5000/100 = 50. This implies that every 50th encounter from the book would be selected.

These sampling procedures control for seasonal effects as well as for different times of the day. However, any reasonable alternative procedure for accomplishing the same distribution of cases is acceptable, provided it is clearly specified and followed in a similar way by all data collectors.

Step 4 (OPTIONAL): Select alternate encounters and link to other data

Sometimes not all the required data will be contained in the chronological listings from which a retrospective sample is drawn. Another type of record may need to be linked to identify the drugs prescribed for the encounter selected, e.g. medical records or pharmacy prescribing slips. This linkage can usually be made on the basis of patient name or identification number, and date. In such situations, where 100% successful linkage may not be possible, it is advisable to select an alternate encounter from the day in question to include in the sample when the primary encounter is not found. This alternate encounter should be specified, e.g. as the next case recorded in the treatment log.

It is usually more efficient to make a complete listing of the encounters to be included in the sample before trying to link to the second data source. When the actual prescribing information is recorded, an alternative encounter is only included in the sample if the primary encounter for that day cannot be found.

Prospective methods for sampling encounters

There are a number of situations where prospective sampling is needed to measure the drug use indicators. Historical drug prescribing data may not exist in health facilities, or the quality and accessibility of the data might be low. To learn about prescribing in such situations, it is preferable to record data for prospective cases as they present for treatment. The following section describes how to implement prospective sampling in an efficient and standardized way.

Step 1: Plan the logistics of obtaining the necessary information

In order to plan how data on prescribing encounters are to be collected, it is necessary to have a basic understanding of the daily procedures in the selected facility. The goal at this stage is to design the most practical prospective system for collecting encounter data. Particular questions that need to be answered are:

• What is the easiest way to identify patients who qualify for the study? Is it easiest to interview patients as they arrive? Is it less disruptive or more efficient to wait until they emerge from the treatment room, or even until they leave the health facility?

• What is the most efficient way to find out the type and quantity of drugs prescribed? Sometimes written information on drugs and quantities to be dispensed is carried by the patient from the treatment room to the dispensary. However, in some settings, prescribed drugs are dispensed in standard quantities by dispensary personnel; sometimes even the decision on which drugs to dispense for certain problems is made in advance, and dispensary attendants distribute a standard package of drugs for a given diagnosis. In such cases, data on which drugs and how much of each drug a patient will receive are not available until after the pharmacy visit has ended.

• If patient treatment cards or dispensed drugs are to be examined, where is the least disruptive location to intercept patients? Often the presence of outsiders collecting information in a clinic may not only interfere with patient flow, but can also make health workers very sensitive to their own behaviour. To reduce disruption, it helps to locate an area separate from the rest of the clinic activities in which to interview already treated patients.

• How long will it take to collect the desired number of cases? Are there certain facilities where data collection will need to be organized in a different way? Is the sample size feasible within existing financial and time constraints? If not, can sample size or study objectives be adjusted so that the study is feasible?

Step 2: Decide who will collect the data

There are two basic options for collecting prescribing data in a prospective study, which can be used alone or in combination. Specially trained data collectors may visit each site and remain there to record data on the planned number of patient encounters as they present for care. Alternatively, prescribers, dispensers or other staff already present at each site are trained to record the data on prescribing encounters and are provided with clear guidelines about which cases and how many cases are to be included in the sample.

The decision about the balance between these two methods of data collection involves a number of trade-offs. There is a cost involved in sending staff to each site to collect data, especially if the frequency of patient visits is low and the number of cases to be sampled is relatively high. If it will only take one or two days to collect all necessary data, or if transport and lodging costs are not excessive, using special data collectors is likely to result in more reliable data.

On the other hand, having facility staff collect the data offers the long-term prospect of actively involving them in quality improvement. Becoming more aware of their own prescribing practices can be one vehicle for improving drug use. However, having to collect and record prescribing data adds one more task to the duties of health staff, and it may be difficult to implement such a process in a reliable way.

There is potential bias in both methods. Staff may react to the presence of outsiders by altering their practices to conform more with a perceived norm. However, the danger of this bias may be even greater if staff at the facility collect data about their own behaviour.

If the goal is to establish a regular monitoring system, the best option may be to implement a system in which data are recorded and reported by facility staff with some mechanism for establishing their validity, perhaps through random site visits by supervisory staff.

For a study in which data need to be collected for a number of days, one possibility is for data collectors to record cases for one or two days, while training local staff in the use of prescribing indicator forms. Forms for the remaining cases can then be left at the facility for local staff to complete and return to the study office. Results for the cases collected by data collectors can then be compared with results from the forms completed after they left, to check for gross differences in key indicators that may be subject to reporting bias, e.g. the percent of patients receiving antibiotics or injections.

Step 3: Design the data collection process

Once a decision has been made about who will collect prescribing data, the process they will use to record the data needs to be clearly specified in order to ensure that data collected in different facilities are comparable. This data collection plan should be based on a pilot test of the logistics of collecting data at a number of sample facilities.

If trained data collectors are used, the procedures to be followed can be taught during pre-study training. The data collection plan should specify: where data collectors should position themselves in order to disrupt the process of clinical care and dispensing as little as possible; how to identify the drugs prescribed (pharmacy record, observation of actual drugs received, etc.); and what to do when the number of patients is less than required.

If local staff are to record encounters prospectively, or if a combination of special data collectors and local staff is to be used, the data collection plan should describe: who should be responsible for recording the data; how they should determine whether or not a given patient is to be included in the sample, e.g. spreading cases over time; and the size of the sample.

Step 4: Select patient encounters

The selection of patient encounters is somewhat different for the groups of core indicators, and depends also on the lay out and procedures of the health facility. For example, the prescribing information (indicators 1-5) could be collected for 30 consecutive encounters if the data collector sits with the prescriber, or just outside the treatment room. In this case it seems advisable to take the 30 encounters somewhere in the middle of the clinic day. Information on consultation time (indicator 6) can then also be clocked. Preferably, patient encounters should be spread randomly over the day, but this may not always be practical. If more prescribers are working simultaneously, a reasonable number of encounters should be taken from each.

Information on the other patient care indicators (7-10) is to be collected somewhere in or around the dispensing area or pharmacy. Dispensing time can easily be measured for a number of consecutive patients. The time needed to collect the information on the other three indicators will determine whether consecutive cases can be recorded or whether it will be necessary to skip patients who have passed through the system in the meantime.

Information on indicators 11 and 12 is facility specific and does not need prospective sampling.