Last observation carried forward LOCF

One very simplistic way of handling missing data is to remove those patients with missing data from the analysis in a complete cases analysis or completers analysis. By definition this will be a per-protocol analysis which will omit all patients who do not provide a measure on the primary endpoint and will of course be subject to bias. Such an analysis may well be acceptable in an exploratory setting where we may be looking to get some idea of the treatment effect if every subject were to follow the protocol perfectly, but it would not be acceptable in a confirmatory setting as a primary analysis. 

This analysis takes the final observation for each patient and uses it as that patient s endpoint in the analysis. For example, in a 12 month trial in acute schizophrenia, a patient who withdraws at month 7 due to side effects will have their month 7 value included in the analysis of the data. 

In one sense this approach has clinical appeal. The final value provided by the patient who withdrew at month 7 is a valid measure of how successful we have been in treating this patient with the assigned treatment and so should be part of the overall evaluation of that treatment. In some circumstances, however. 

A success/failure approach will be particularly effective if the endpoint is already binary, for example cured/not cured in a trial of an anti-infective. In the earlier discussion of modified ITT the CPMP (1997) Note for Guidance on the Evaluation of New Anti-bacterial Medicinal Products goes on to say Patients who have no measurements after baseline are included as failures in the analysis.  

This method gives those subjects who withdraw for positive reasons the best possible outcome value for the endpoint and those who withdraw for negative reasons, the worst value. This may seem a little extreme and a lesser position would be to look at the distribution of the endpoint for the completers and to use say the upper quartile (the value that cuts off the best 25 per cent of values) for those subjects withdrawing for positive reasons and the lower quartile (the value that cuts off the worst 25 per cent of values) for those subjects who withdraw for negative reasons. 

---

Key Concepts for Creating Analysis Data Sets 84 Defining Variables Once 84 Defining Study Populations 85 Defining Baseline Observations 85 Last Observation Carried Forward (LOCF) 86 Defining Study Day 89 Windowing Data 91 Transposing Data 94... 

Program 4.1 Deriving Last Observation Carried Forward (LOCF) Variables... 

Clinical assessments were made by specialized raters not involved in the treatment after 1, 4, 8, 12 and 16 weeks. The study was intended to include 96 patients, 15 of whom withdrew at the very beginning when they heard what treatment they were to be given. Most of those withdrawing were in the psychotherapy group. In addition, patients who failed to show any pronounced improvement in their symptoms within 8 weeks or whose condition even deteriorated were withdrawn from the study. In the case of those patients withdrawing from the study, scores recorded at the time of withdrawal were used for purposes of evaluation this represents a last observation carried forward (LOCF) analysis. Withdrawals from the study occurred rather frequently and showed the following distribution ... 

In either case, reaching this point indicates that the drug is beneficial or not and is at least a qualitative endpoint. Last observation carried forward (LOCF), a standard method of data analysis, carries the last data point forward week by week. Random regression models can estimate what would happen at a later time point, assuming that patients change in a linear fashion. Improvement, however, often levels off. Thus, creating data points based on questionable assumptions can potentially introduce substantial bias. 

An international, multicenter, double-blind trial addressed the acute efficacy and safety of a single-dose range of olanzapine (5 to 20 mg/day) compared with a single-dose range of haloperidol (5 to 20 mg/day) (11.6). A total of 1996 patients with a DSM-lll-R diagnosis of schizophrenia (83.1%), schizophreniform disorder (1.9%), or schizoaffective disorder (15%) participated in this study. The primary overall efficacy analysis (i.e., the difference in baseline to endpoint (last observation carried forward [LOCF]) mean change on the BPRS) found olanzapine to be statistically superior to haloperidol (HPDL) (i.e., -10.98 -7.93 p < 0.015). 

The last of these approaches is called imputation of missing values. As Piantadosi (2005) commented, while this approach sounds a lot like making up data, when done properly it may be the most sensible strategy. While techniques for addressing missing data can be technically difficult, one commonly used, simple imputation method is called last observation carried forward (LOCF). In a study with repeated measurements over time, the most recent observation replaces any subsequent missing observations (Piantadosi, 2005, see also Molenberghs and Kenward, 2007). 

For example, in the analgesic example cited above (25), a comparison was made between an analysis using the last observation carried forward (LOCF) method and the proposed mixed effects maximum likelihood method. Although this was a retrospective analysis, similar contrasts could be included in the trial s simulation to ascertain the most appropriate analytical methodology to include in the study design (protocol). Other analysis factors for consideration include appropriate correction of variability, where such sources may include differences between sites or regional differences. 

Last observation carried forward (LOCF) analysis is the standard approach with a... 

---