Missing Data in Systematic Reviews: When and How to Contact Study Authors

Missing information creates both an analytical gap and a methodological choice

A report can establish that a study is eligible yet fail to provide the information required for synthesis. Dispersion may be absent, denominators may be unclear, outcomes may be reported only graphically, or intervention details may not support classification. The review team must decide whether to contact investigators, derive a value, use an alternative result, impute under a declared rule, exclude the observation or retain the study only in narrative synthesis. Each option changes uncertainty and can influence which evidence contributes.

Author contact is sometimes treated as an obviously superior solution, but it has limitations. Records may no longer exist, contact details may be obsolete, investigators may lack time, and responses may differ across studies. An evaluation of systematic reviews obtained additional data from 28 of 45 contacted authors while many requested items remained unresolved.¹² A Cochrane review of methods for obtaining unpublished data also illustrates variation in strategies and outcomes.¹¹ These findings support planned, transparent contact; they do not establish a universal response rate or ideal follow-up schedule.

PRISMA 2020 asks review authors to report processes for obtaining or confirming data from investigators.^2,3 Transparent reporting begins with a complete internal event history. If the request, response and interpretation are not recorded in your Systematic review data extraction form template or dedicated contact log, the analysis can contain information that readers and future reviewers cannot reconstruct. The published method may summarize the process, but the log should retain study ID, field, source checks, dates, channel, request, responses, files, reviewer interpretation and analytical status.

Predefine the contact policy before difficult cases accumulate

The protocol or data-management plan should state the purposes for which contact will be considered. Possible purposes include confirming study identity, clarifying eligibility, obtaining outcome statistics, resolving denominator or unit ambiguity, understanding methods, or requesting unpublished aggregate results. The policy should distinguish essential fields from optional descriptive details so workload follows analytical importance.

Define which sources must be checked first, who may contact authors, the approved channel, how identity is confirmed, whether a follow-up is planned, the review-specific stopping rule and the action after non-response. Because fields and disciplines differ, no single number of attempts or interval should be imposed as an Academy-wide methodological fact. The chosen procedure should be reasonable for the study age, urgency, burden and communication channel and should be applied consistently to comparable cases.

Plan how new information will be stored, verified and cited. Author-supplied files may contain unpublished material or personal information, so use institutional storage and appropriate access. Decide how the manuscript will acknowledge assistance and whether permission is needed for quoted or shared material. These governance choices should be settled before a response arrives.

Triage the gap by necessity, recoverability and consequence

Begin by defining the missing field precisely. “Variance missing” is incomplete if the study reports several outcomes, arms and time points. State the outcome, measure, analysis population, time, arm and statistic. Confirm that all linked reports, supplements, registries, repositories and corrections have been searched. Sometimes the data are present under a different analysis or in a companion publication.

Next ask whether the field is necessary. A value required for the primary synthesis has a different priority from a detail that would merely enrich a descriptive table. Consider whether a prespecified alternative is available and whether contact could change eligibility or analysis. Prioritisation should not be based on whether a result appears favourable; outcome-blind rules reduce the risk of selective pursuit.

Finally consider whether the request is answerable and proportionate. Authors are more likely to respond accurately to a concise question that identifies the study and the exact information required. Requests for reconstructed datasets many years after publication may have low feasibility and high burden. This does not make contact inappropriate, but it should influence expectations, timelines and the declared fallback.

Construct the request as a reproducible query

The message should identify the review team and institutional context, cite the study unambiguously, and explain why the information is needed. State what has already been found and ask the exact question. If requesting a statistic, specify arm, outcome, time point, analysis population, unit and preferred format. If a calculation could be made from several inputs, ask for those inputs rather than an unexplained transformed value.

Use neutral language. Do not imply that a particular answer is expected or that authors must defend their work. Invite correction of the review team’s understanding. Acknowledge that records may be unavailable and that a brief confirmation can still be useful. Keep the number of questions manageable and consider a structured table when several closely related fields are needed.

Protect information in transit. Do not request participant-identifiable data through ordinary email. For files or individual-participant data, use a secure institutional process with data-sharing, ethics and governance arrangements as required. Aggregate clarification can often be handled with less infrastructure, but it still belongs in controlled storage.

Record correspondence as events, not as one final status

A contact history may contain a drafted request, sent message, delivery failure, follow-up, partial response, clarification, file transfer and final review decision. Each event should have a date, channel, sender or responsible reviewer, status and link to the relevant study and field. Do not overwrite “sent” with “resolved”; the sequence matters for reporting and updates.

The log should also preserve the applicable dataset and form version. A response received after analysis freeze may support an update but not the published estimate. If the value is incorporated, create a new data version and record affected calculations. A later correction from the author should append another event and should not erase the earlier response.

Personal email addresses can be stored separately from the methodological log. The reader-facing or shared corpus usually needs the process and evidence state, not contact details. Separation supports privacy while retaining auditability.

Appraise what the response actually contributes

Classify the response before extracting a number. A clarification explains information already reported. A correction changes the publication record. Newly supplied aggregate data add evidence not present in the report set. A statement that records are unavailable documents feasibility but does not establish the missing value. A partial or ambiguous reply may leave the query unresolved. These categories have different implications for provenance and reporting.

Confirm that the respondent and material can be linked to the correct study. Check whether the arm, population, outcome, time, statistic, unit and denominator match the request. Compare the value with published totals and related results. An apparent standard deviation may be a standard error; an overall result may not be arm specific; a denominator may reflect complete cases. Ask a focused clarification if the original policy permits, otherwise retain the unresolved state.

Author authority does not remove the need for review. Investigators can make mistakes, interpret an old question differently or provide a value from an analysis that does not match the protocol. Preserve the response and the review team’s interpretation as distinct records. High-consequence additions should receive independent verification and reproducible derivation where applicable.

Integrate author-supplied information without erasing provenance

Store the published value or missing state, the author-supplied information, the review decision and the analysis value separately. Identify the source status in the analysis dataset. If the response corrects a publication, retain both values and record why the corrected value was used. If it supplies inputs for a derivation, retain the inputs, formula or code and output.

Do not let the presence of a response determine inclusion automatically. The value must still satisfy the protocol’s outcome, time and analysis rules. Conversely, a clarification can resolve eligibility or classification without supplying a numerical result. The log should connect each response to the exact decision it affects.

At analysis freeze, enumerate author-supplied values, unresolved contacts and responses excluded from use. This enables sensitivity analyses that compare source-status groups when justified and permits the manuscript to describe the contribution of contact accurately.

Treat non-response as unresolved availability, not as a study characteristic

Failure to receive a reply does not show that data are absent, that the authors are unwilling to share, or that the result has a particular direction. Delivery can fail; investigators can change institutions; records can be archived; requests can arrive during leave. Close the event under the prespecified rule and apply the planned analytical treatment without inferring motive.

Selective success can nevertheless affect the evidence base. Studies with current contact details, recent publication, certain sponsors or particular results may be more likely to provide data. Describe how many studies and fields were contacted, the response categories and how supplied data were used. Consider sensitivity analysis or limitation statements when contact materially changes which studies contribute.

A response rate alone is insufficient because one study can receive several requests and a response can resolve none, some or all fields. Use both study-level and query-level summaries where informative. Define the denominator: authors approached, studies contacted, messages delivered or data items requested are not interchangeable.

Report the actual method and its analytical consequences

The methods should state why investigators were contacted, the sources checked first, who made contact, the communication and follow-up procedure, and how responses were verified and incorporated. Results should summarize contacts and outcomes and identify material changes to the evidence. The discussion should address unresolved data and the possibility of selective availability.

Do not report “authors were contacted where necessary” if necessity was never defined. A concise manuscript can refer to a protocol or supplement for the full policy. PRISMA reporting should reflect what happened, including deviations. If no contact was performed because the review used a prespecified alternative, state that method rather than implying that authors failed to respond.

When correspondence supplies data used in a published analysis, follow journal and institutional policies on acknowledgement, citation and data availability. Do not publish private correspondence or files without appropriate permission. Transparency concerns the method and evidence status, not unrestricted disclosure.

Design the contact record for updates and correction

Future reviewers should be able to see which questions were asked, which addresses or channels failed, what responses were received and whether a later reply arrived. Stable study and query IDs prevent repeated burden and help an update team distinguish a still-open query from a field resolved under a previous analysis.

Contact details may require refresh and should be retained only as long as justified by policy. Methodological event history can remain after personal contact data are removed or separated. Store attachments using controlled filenames and checksums where appropriate, and record access restrictions.

If a response changes a published conclusion, follow the review’s correction and versioning process. The contact log provides evidence for what changed and when; it is not itself a publication correction. Connect the new data version, analysis output and reporting action.

Know when further contact is unlikely to resolve the decision

Repeated messages cannot compensate for an ill-defined question, an unavailable historical record or an outcome that was never measured. If the response indicates that the requested data do not exist, or if the supplied information cannot be linked to the required population and time, close the query under the planned rule. Continuing contact without a plausible route to resolution increases burden and may create inconsistent treatment across studies.

Escalate scientific ambiguity inside the review team. An author can clarify what was done, but the review must decide whether that method fits its protocol and estimand. Separate factual clarification from the review’s eligibility or analytical judgment and record both.

Conclusion: author contact is a governed extension of data extraction

A defensible contact process begins with a precise analytical need, searches existing sources, asks an answerable and neutral question, preserves every event, appraises the response and records how the evidence changed. It applies a declared rule after non-response and acknowledges the possibility of selective availability. No template can guarantee access to records or validate an author-supplied value automatically.

The paired Missing data contact log and request templates operationalise this event history. Adapt the policy, privacy controls, timeline and message to the review context. The final dataset should always distinguish published, author-supplied, derived and unresolved information so that the synthesis remains auditable.

References

1. Li T, Higgins JPT, Deeks JJ. Chapter 5: Collecting data. In: Higgins JPT, Thomas J, Chandler J, Cumpston M, Li T, Page MJ, Welch VA, editors. Cochrane Handbook for Systematic Reviews of Interventions. Version 6.5. Cochrane; 2024. Access the current chapter.
2. Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021;372:n71. doi:10.1136/bmj.n71.
3. Page MJ, Moher D, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. PRISMA 2020 explanation and elaboration: updated guidance and exemplars for reporting systematic reviews. BMJ. 2021;372:n160. doi:10.1136/bmj.n160.
4. Aromataris E, Lockwood C, Porritt K, Pilla B, Jordan Z, editors. JBI Manual for Evidence Synthesis. Adelaide: JBI; 2024. doi:10.46658/JBIMES-24-01.
5. Jonnalagadda SR, Goyal P, Huffman MD. Automating data extraction in systematic reviews: a systematic review. Syst Rev. 2015;4:78. doi:10.1186/s13643-015-0066-7.
6. Moher D, Shamseer L, Clarke M, Ghersi D, Liberati A, Petticrew M, et al. Preferred reporting items for systematic review and meta-analysis protocols (PRISMA-P) 2015 statement. Syst Rev. 2015;4:1. doi:10.1186/2046-4053-4-1.
7. Higgins JPT, Thomas J, Chandler J, Cumpston M, Li T, Page MJ, Welch VA, editors. Cochrane Handbook for Systematic Reviews of Interventions. 2nd ed. Chichester: Wiley; 2019.
8. Mathes T, Klaßen P, Pieper D. Frequency of data extraction errors and methods to increase data extraction quality: a methodological review. BMC Med Res Methodol. 2017;17:152. doi:10.1186/s12874-017-0431-4.
9. Young T, Hopewell S. Methods for obtaining unpublished data. Cochrane Database Syst Rev. 2011;(11):MR000027. doi:10.1002/14651858.MR000027.pub2.
10. Selph SS, Ginsburg AD, Chou R. Impact of contacting study authors to obtain additional data for systematic reviews: a diagnostic accuracy study. Syst Rev. 2014;3:107. doi:10.1186/2046-4053-3-107.